2014 IGES Scientific Meeting Abstracts

Transcription

ABSTRACTSFROMTHE23RDANNUAL
MEETINGOFTHEINTERNATIONALGENETIC
EPIDEMIOLOGYSOCIETY
Vienna,Austria
28‐30August2014
ISBN:978‐1‐940377‐01‐8
Abstractsfromthe23rdAnnualMeetingoftheInternational
GeneticEpidemiologySociety,Vienna,Austria
August28‐30,2014
ScientificProgramCommittee
JustoLorenzoBermejo(Chair)
UniversityHospitalHeidelberg
Heidelberg,Germany
(2013‐2015)
CeliaGreenwood
McGillUniversity
Montreal,Canada
(2012‐2014)
JeanineHouwing‐Duistermaat
LUMC
Leiden,TheNetherlands
(2012‐2014)
AndrewDPaterson
TheHospitalforSickChildrenResearchInstitute
Toronto,Canada
(2013‐2015)
AndréScherag
CSCC,UniversityHospitalJena
Jena,Germany
(2014‐2016)
NathanTintle
DordtCollege,SiouxCenter
Iowa,USA
(2014‐2016)
FranceGagnon(President‐Elect)
UniversityofToronto
Toronto,Canada
AlexanderFWilson(President)
NIH/NHGRI
Baltimore,MDUSA
Dr.Wilsonisservinginhispersonalcapacity.
ThisvolumewasformattedbyBarbaraPeil,ÖzgeKaradag,RosaGonzálezSilos,MariaKabischand
CarineLegrand,PhDstudentsattheStatisticalGeneticsGroup,InstituteofMedicalBiometryand
Informatics,UniversityofHeidelberg,Germany
Mini‐Symposium
M1 Applicationofgenomictestsinbreastcancermanagement
M2 Riskpredictionmodelsusingfamilyandgenomicdata
M3 Theimportanceofappropriatequalitycontrolin‐omicsstudiesasrequiredfor
personalizedandstratifiedmedicine
M4 Studydesignsforpredictivebiomarkers
EducationalWorkshop
E1
Pharmacogenomics:past,presentandfuture
E2
Assessingthegeneticbasisofdrugresponse
E3
Clinicalutilityinpharmacogenomics:gettingbeyondindividualvariants
E4
Smokingbehaviorandlungcancerriskrelatedtonicotinicacetylcholinereceptorvariants
andmetabolicvariants
InvitedSpeakers
I1
Enrichmentdesignsforthedevelopmentofpersonalizedmedicine
I2
Causalassociationstructuresin‐omicsdata:howfarcanwegetwithstatisticalmodeling?
I3
Therelevanceofepigenomicsforpersonalizedmedicine
I4
Finemappingofcomplextraitlociwithcoalescentmethodsinlargecase‐controlstudies
I5
Theinterfacehypothesisinexplaininghost‐bacterialinteractionsinthehumangut
NeelandWilliamsAwardCandidates
A1
Anovelmethodusingcrosspedigreesharedancestrytomaprarecausalvariantsinthe
presenceoflocusheterogeneity
A2
Survivalanalysiswithdelayedentryinselectedfamilieswithapplicationtohumanlongevity
A3
Combiningfamily‐andpopulation‐basedimputationdataforassociationanalysisofrare
andcommonvariantsinlargepedigrees
A4
Mixedmodelingfortime‐to‐eventoutcomeswithlarge‐scalepopulationcohortsand
genome‐widedata
A5
Thecollapsedhaplotypepatternmethodforlinkageanalysisofnext‐generationsequencing
data
A6
Meta‐analysisapproachforhaplotypeassociationtests:ageneralframeworkforfamilyand
unrelatedsamples
ContributedPlatformPresentations
C1 Identificationofbloodpressure(BP)relatedcandidategenesbypopulation‐based
transcriptomeanalyseswithintheMetaXpressconsortium
C2 Mixed‐modelanalysisofcommonvariationrevealspathwaysexplainingvarianceinAMD
risk
C3 Aphenome‐wideassociationstudyofnumerouslaboratoryphenotypesinAIDSclinical
trialsgroup(ACTG)protocols
C4 eMERGEphenome‐wideassociationstudy(PheWAS)identifiesclinicalassociationsand
pleiotropyforfunctionalvariants
C5
C6
C7
C8
C9
C10
C11
C12
C13
C14
C15
C16
C17
C18
C19
C20
C21
AnovelG‐BLUP‐likephenotypepredictorleveragingregionalgeneticsimilarityandits
applicationsinpredictingdiseaseseverityanddrugresponse
MitochondrialGWAanalysisinseveralcomplexdiseasesusingtheKORApopulation
AdramaticresurgenceoftheGIGOsyndromeinthe21stcentury
Largescalepredictionanddissectionofcomplextraits
Geneticpredictorsoflongertelomeresarestronglyassociatedwithriskofmelanoma
DetectionofcisandtranseQTLs/mQTLsinpurifiedprimaryimmunecells
Whynext‐generationsequencingstudiesmayfail:challengesandsolutionsforgene
identificationinthepresenceoffamiliallocusheterogeneity
Variationinestimatesofkinshipobservedbetweenwhole‐genomeandexomesequence
data
Robustgenotypecallingfromverylowdepthwholegenomesequencingdata
Insightsintothegeneticarchitectureofanthropometrictraitsusingwholegenomesequence
data
Standardimputationversusgeneralizationsofthebasiccoalescenttoestimategenotypes
Improvementofgenotypeimputationaccuracythroughintegrationofsequencedatafroma
subsetofthestudypopulation
Learninggeneticarchitectureofcomplextraitsacrosspopulations
Genome‐widegenotypeandsequence‐basedreconstructionofthe140,000yearhistoryof
modernhumanancestry
Modelcomparisonandselectionforcountdatawithexcesszerosinmicrobiomestudies
Bayesianlatentvariablemodelsforhierarchicalclusteredtaxacountsinmicrobiomefamily
studieswithrepeatedmeasures
Aretrospectivelikelihoodapproachforefficientintegrationofmultipleomicsandnon‐
omicsfactorsincase–controlassociationstudiesofcomplexdiseases
Inferenceforhigh‐dimensionalfeatureselectioningeneticstudies
C22
Posters
P1
Increasedpowerfordetectionofparent‐of‐origin(imprinting)effectsingenome‐wide
associationstudiesusinghaplotypeestimation
P2
EpidemiologicalProfileofCleftPalateintheStateofBahia‐Brazil
P3
GeneralizedFunctionalLinearModelsforGene‐basedCase‐ControlAssociationStudies
P4
Geneticanalysisofthechromosome15q25.1regionidentifiesIREB2variantsassociated
withlungcancer
P5
Anovelintegratedframeworkforlargescaleomicsassociationanalysis
P6
InclusiveCompositeIntervalMappingandSkew‐NormalDistribution
P7
Transmission‐basedTestsForGeneticAssociationUsingSibshipData
P8
Identificationofrarecausalvariantsinsequence‐basedstudies
P9
TargetedresequencingofGWASloci:insightintogeneticetiologyofcleftlipandpalate
throughanalysisofrarevariantswithfocusonthe8q24region
P10 AjointassociationmodelofeffectsofrareversuscommonvariantsonAge‐relatedMacular
Degeneration(AMD)usingaBayesianhierarchicalgeneralizedlinearmodel
P11 AssociationBetweenBloodPressureSusceptibilityLociandUrinaryElectrolytes
P12
P13
P14
P15
P16
P17
P18
P19
P20
P21
P22
P23
P24
P25
P26
P27
P28
P29
P30
P31
P32
P33
P34
P35
P36
P37
P38
P39
Asystematicevaluationofshorttandemrepeatsinlipidcandidategenes:ridingontheSNP‐
wave
Linkagedisequilibriummappingofmultiplefunctionallociincase‐controlstudies
Geneticvariantsintransporterandmetabolizinggenesandsurvivalincolorectalcancer
patientstreatedwithoxaliplatincombinationchemotherapy
Post‐Genome‐WideAssociationStudyUsingGeneralizedStructuredComponentAnalysis
DetectingMaternal‐FetalGenotypeInteractionsAssociatedwithConotruncalHeartDefects:
AHaplotype‐basedAnalysiswithPenalizedLogisticRegression
Mutationsscreeningofexons7and13ofTMC1gene(DFNB7/11)inIranianautosomal
recessivenon‐syndromichearingloss(NSHL)probandsusingmoleculartechniques
ConotruncalHeartDefectsandCommonVariantsinMaternalandFetalGenesinFolate,
HomocysteineandTranssulfurationPathways
GeneticPredispositionofXRCC1inSchizophreniaPatientsofSouthIndianPopulation
Astochasticsearchthroughsmokingimagesinmovies,geneticandpsycho‐socialfactors
associatedwithsmokinginitiationinMexicanAmericanyouths
AssociationbetweenApolipoproteinEgenotypeandcancersusceptibility:ameta‐analysis
NovelapproachidentifiesSNPsinSLC2A10andKCNK9withevidenceforparent‐oforigin
effectonbodymassindex
InteractiveeffectbetweenDNAH9geneandearly‐lifetobaccosmokeexposureinbronchial
hyper‐responsiveness
DetectionofrarehighlypenetrantrecessivevariantsusingGWASdata
CopyNumberVariation(CNV)detectioninwholeexomesequencingdataforMendelian
disorders
Combininggeneticandepigeneticinformationidentifiedimprinted4q35variantassociated
withthecombinedasthma‐plus‐rhinitisphenotype
Bayesianlatentvariablecollapsingmodelfordetectingrarevariantinteractioneffectintwin
study
RareVariantAssociationTestforNuclearFamilies
Samplesizeandpowerdeterminationforassociationtestsincase‐parenttriostudies
IntegrationofDNAsequencevariationandfunctionalgenomicsdatatoinfercausalvariants
underlyingchemotherapeuticinducedcytotoxicityresponse
LargeScalePredictionandDissectionofComplexTraits
ImputationforSNPsusingsummarystatisticsandcorrelationbetweengenotypedata
EvaluationofpopulationstratificationinalargebiobanklinkedtoElectronicHealthRecords
Estimatinggeneticeffectsonsusceptibilityandinfectivityforinfectiousdiseases
CombinedMethodstoExploreGeneticEtiologyofRelatedComplexDiseases
Integrativeanalysisofsequencingandarraygenotypedatafordiscoveringdisease
associationswithraremutations
Amethodforfastcomputationoftheproportionofvariantsaffectingacomplexdiseaseand
oftheadditivegeneticvarianceexplainedinGWASSNPstudies.
Correctingforsampleoverlapincross‐traitanalysisofGWAS
Epigenome‐wideassociationstudyofcentralizedadiposityin2,083AfricanAmericans:The
AtherosclerosisRiskinCommunities(ARIC)Study
P40
P41
P42
P43
P44
P45
P46
P47
P48
P49
P50
P51
P52
P53
P54
P55
P56
P57
P58
P59
P60
P61
P62
P63
P64
P65
P66
P67
Canlow‐frequencyvariantsberescuedingenome‐wideassociationstudiesusingsparse
datamethods?
Anovelkernel‐basedstatisticalapproachtotestingassociationinlongitudinalgenetic
studieswithanapplicationofalcoholusedisorderinaveterancohort
AGene‐EnvironmentInteractionBetweenCopyNumberBurdenandOzoneExposurein
RelationtoRiskofAutism
Choosingacase‐controlassociationteststatisticforlow‐countvariantsintheUKBiobank
LungExomeVariantEvaluationStudy
SNPcharacteristicspredictreplicationsuccessinassociationstudies
Data‐DrivenWeightedEncoding:ANovelApproachtoBiallelicMarkerEncodingfor
EpistaticModels
AOne‐Degree‐of‐FreedomTestforSupra‐MultiplicativityofSNPEffects
Fine‐mappingeGFRsusceptibilitylocithroughtrans‐ethnicmeta‐analysis
Arelipidriskallelesidentifiedingenome‐wideassociationstudiesreadyfortranslationto
clinicalstudies?
Genome‐widemeta‐analysisofsmoking‐dependentgeneticeffectsonobesitytraits:the
GIANT(GeneticInvestigationofANthropometricTraits)Consortium
ABinomialRegressionModelforAssociationMappingofMultivariatePhenotypes
HowtoincludechromosomeXinyourgenome‐wideassociationstudy
Exomechipmeta‐analysistoidentifyrarecodingvariantsassociatedwithpulsepressure
Genome‐widesearchforage‐andsex‐dependentgeneticeffectsforobesitytraits:Methods
andresultsfromtheGIANTConsortium
Meta‐analysisofgene‐setanalysesbasedongenomewideassociationstudies
Meta‐analysisofcorrelatedtraitsusingsummarystatisticsfromGWAS
StudyingtheEthnicDifferencesintheGeneticsofType2DiabetesusingthePopulation
SpecificHumanPhenotypeNetworks
HierarchicalBayesianModelintegratingsequencingandimputationuncertaintyusing
MCMCmethodforrarevariantassociationdetection
Sex‐specificassociationofMYLIPwithmortality‐optimizedhealthyagingindex
Geneticdeterminantsofliverfunctionandtheirrelationshiptocardio‐metabolichealth
VariableselectionmethodforcomplexgeneticeffectmodelsusingRandomForests
Identificationofsharedgeneticaetiologybetweenepidemiologicallylinkeddisorderswith
anapplicationtoobesityandosteoarthritis
InvestigationofgeneticriskfactorsofverylowbirthweightinfantswithintheGerman
NeonatalNetwork
Artificialintelligenceanalysisofepistasisinagenome‐wideassociationstudyofglaucoma
Mutationscausingcomplexdiseasemayundercertaincircumstancesbeprotectiveinan
epidemiologicalsense
Genome‐wideAssociationStudyIdentifiesSNPrs17180299andMultipleHaplotypeson
CYP2B6,SPON1andGSG1LAssociatedwithPlasmaConcentrationsoftheMethadoneR‐and
S‐enantiomerinHeroin‐dependentPatientsunderMethadoneMaintenanceTreatment
Anonparametricregressionapproachtotheanalysisofgenomewideassociationstudies
Geneticinsightsintoprimarybiliarycirrhosis–aninternationalcollaborativemeta‐analysis
andreplicationstudy
P68
P69
P70
P71
P72
P73
P74
P75
P76
P77
P78
P79
P80
P81
P82
P83
P84
P85
P86
P87
P88
P89
P90
P91
P92
P93
P94
P95
P96
GenesAssociatedwithLungCancer,ChronicObstructivePulmonaryDisease,orBoth
Ageneralapproachforcombiningdiverserarevariantassociationtestsprovidesimproved
poweracrossawiderrangeofgeneticarchitecture
AMethodologicalComparisonofEpistasisModelingofHighOrderGene‐GeneInteractions
withApplicationtoGeneticProfilingofPAInfectionamongCysticFibrosisPatients
eQTLandpathwayanalysisonexpressionprofilesofacattlecross
Evidenceforpolygeniceffectsintwogenome‐wideassociationstudiesofbreastcancer
usinggeneticallyenrichedcases
DoBoundariesMatterforTiledRegression?
METAINTER:meta‐analysistoolformultipleregressionmodels
SuccessfulreplicationofGWAShitsformultiplesclerosisin10,000Germansusingthe
exomearray
SharedGeneticEffectsUnderlyingAgeatMenarche,AgeatNaturalMenopauseandBlood
Pressure
IdentificationofcombinedCommon‐andRare‐Geneticvariancesassociatedwithrenal
functioninHanChinese
Pathwayandgene‐geneinteractionanalysisrevealsnewcandidategenesformelanoma
Leveragingevolutionarilyconserved,celltype‐specific,regulatoryregiondatatodetect
novelSNP‐TFPIassociations
Asoftwarepackageforgenome‐wideassociationstudieswithRandomSurvivalForests
Identificationofnovelcommonandraregeneticvariantsassociatedwithrenalfunctionin
HanChinese
AGenome‐WideAssociationStudytoExploreGene‐environmentInteractionwithParental
SmokingandtheRiskofChildhoodAcuteLymphocyticLeukemia
Network‐basedanalysisofGWASdata:Doesthegene‐wiseassociationsignificance
modelingmatters?
Heritabilityestimatesandgeneticassociationfor60+complextraitsinayounghealthy
siblingcohort
Large‐scaleexomechipgenotypingrevealsnovelcodingvariationassociatedwith
endometriosis
DissectingtheObesityDiseaseLandscape:IdentifyingGene‐GeneInteractionsthatare
HighlyAssociatedwithBodyMassIndex(BMI)
InvestigationofParent‐of‐OrigineffectsinAutismSpectrumDisorders
IntegrativeclusteringofmultiplegenomicdatausingNon‐negativeMatrixFactorization
Toolsforrobustanalysisingenome‐wideassociationstudiesusingSTATA
Developmentofathree‐waymixedmodellingapproachintegratinggeneticandclinical
variablesinanalysisofearlytreatmentoutcomesinepilepsy.
Meta‐analysisoflowfrequencyandrarecodingvariantsandpulmonaryfunction.
UsingPolygeneScoresandGCTAtoIdentifyaSubsetofSNPsthatContributetoGeneticRisk
ChallengingIssuesinGWASofHumanAgingandLongevity
HeritabilityestimatesonHodgkinlymphoma:agenomicversuspopulationbasedapproach
Areweabletoguidetreatmentchoicetoreduceantidepressant‐inducedsexualdysfunction
inmalesusinggenome‐widedatafromrandomisedcontrolledtrials?
Ageneralmethodfortestinggeneticassociationwithoneormoretraits
P98 AGeneralizedSimilarityUtestforMultiple‐traitSequencingAssociationAnalyses
P99 ModelingX‐chromosomedatainRandomForestGeneticAnalysis
P100 EmpiricalBayesScanStatisticsforDetectingClustersofDiseaseRiskVariantsinGenetic
Studies,withApplicationstoCNVsinAutism
P101 Finemappingofchromosome5p15.33regionforlungcancersusceptibilitybasedona
targeteddeepsequencingandcustomAxiomarray
P102 Geneticvariantsininflammation‐relatedgenesandinteractionwithNSAIDuseoncolorectal
cancerriskandprognosis
P103 AssociationanalysisofexomechipdataofPolycysticOvarySyndromeinEstonianBiobank
P104 Amodelforco‐segregationofcriptorchidismandtestiscancerinfamilies
P105 Jointanalysisofsecondaryphenotypes:anapplicationinfamilystudies
P106 Predictionofimprintedgenesbasedonthegenome‐widemethylationanalysis
P107 AddictionandMentalHealthGenesformGenomicHotspotswithDrugableTargets.
P108 Recurrentsharedrarevariantsin9genesdetectedbywholeexomesequencingofmultiplex
oralcleftsfamilies
P109 Evaluationofvariantcallingfromthousandsoflowpasswholegenomesequencing(WGS)
datausingGATKhaplotypecaller
P110 IntegrationoffMRIandSNPsindicatedpotentialbiomarkersforSchizophreniadiagnosis
P111 EWAStoGxE:Arobuststrategyfordetectinggene‐environmentinteractionmodelsforage‐
relatedcataract
P112 RNA‐seqanalysisoflungadenocarcinomarevealsdifferentialgeneexpressioninnonsmoker
andsmokerpatients
P113 UsingrandomforeststoidentifygeneticlinksbetweenAlzheimer’sdiseaseandtype2
diabetes
P114 StudyoFHumanMGPpromotervariantsinCADpatients:FromExperimenttoprediction
P115 Anovelfunctionaldataanalysisapproachtodetectinggenebylongitudinalenvironmental
exposureinteraction
P116 LeveragingFamilyStructurefortheAnalysisofRareVariantsinKnownCancerGenesfrom
WESofAfricanAmericanHereditaryProstateCancer
P117 Associationofbreastcancerrisklociwithsurvivalofbreastcancerpatients
P118 Evidenceofgene‐environmentinteractionsinrelationtobreastcancerrisk,resultsfromthe
BreastCancerAssociationConsortium
P119 Integrationofpathwayandgene‐geneinteractionanalysesrevealbiologicallyrelevantgenes
forBreslowthickness,amajorpredictorofmelanomaprognosis
P120 JAG1polymorphismisassociatedwithincidentneoplasminasouthernChinesepopulation
P121 Epigenome‐widemethylationarrayanalysisrevealsfewmethylationpatterndifferences
betweenhyperplasticpolypsandsessileserratedadenomas/polyps
P122 Theeffectofbileacidsequestrantsontheriskofcardiovascularevents:Ameta‐analysisand
MendelianRandomizationanalysis
P123 MendelianRandomisationstudyofthecausalinfluenceofkidneyfunctiononcoronaryheart
disease
P124 Sharedgeneticriskofmyocardialinfarctionandbloodlipidsusingempiricallyderived
extendedpedigrees:resultsfromtheBusseltonHealthStudy
P125 AnalysisofCase‐Base‐Controldesigns
P126 PolymorphismsinHTR3A,CYP1A2,DRD4andCOMTandresponsetoclozapinein
treatment‐resistantschizophrenia:agene‐geneinteractionanalysis
P127 Jointmodelingoflongitudinalandtime‐to‐eventphenotypesingeneticassociationstudies:
strengthsandlimitations
P128 Perinataldepressionandomega‐3fattyacids:AMendelianrandomisationstudy
P131 AGene‐EnvironmentInteractionBetweenCopyNumberBurdenandOzoneExposure
ProvidesaHighRiskofAutism
P132 GeneRegulatoryNetworkinferenceviaConditionalInferenceTreesandForests
P133 Predictingthegeneticriskforcomplexdiseases:choosingthebestpolygenicriskscorefor
typeIIdiabetes
P134 Epigenome‐wideassociationwithsolublecelladhesionmoleculesamongmonozygotictwins
P135 Genapha/dbASM:webbasedtoolstoinvestigateallele‐specificmethylation
P136 Agene‐basedmethodforanalysisofIllumina450Kmethylationdata
P137 Takeresearchtothenextlevelwithsecondarydataanalyses:Fine‐mappingthespecific
languageimpairmentgene
P138 DetectionofGene‐GeneInteractioninAffectedSibPairsAllowingforParent‐of‐Origin
Effects
P139 StudyDesignsforPredictiveBiomarkers
P140 DoestheFTOgeneinteractwiththesocio‐economicstatusontheobesitydevelopment
amongyoungEuropeanchildren?ResultsfromtheIDEFICSstudy
P141 IdentificationofClustersinNetworkGraphsbyaCorrelation‐basedMarkovCluster
Algorithm
P142 DevelopnovelmixturemodeltoestimatethetimetoantidepressantOnsetofSSRIsandthe
timingeffectsofkeycovariates
P143 Definingrecombinationhotspotblocks:Justhowhotishot?
P144 Complexgenealogies,simplegeometricstructures
P145 Missingheritabilitypartiallyexplainedbysequentialenrollmentofstudyparticipants
P146 RobustPrincipalComponentAnalysisAppliedtoPopulationGeneticsProcesses
P147 Identifyingfoundersmostlikelytohaveintroduceddisease‐causingmutationswiththeR
packageGenLib
P148 RegionalIBDAnalysis(RIA):linkageanalysisinextendedpedigreesusinggenome‐wideSNP
data
P149 Polygenicriskpredictionmodelinginpedigreesimprovespower
P150 Performanceoflinkageanalysisconductedwithwholeexomesequencingdata
P151 Useofexomesequencingdatafortheanalysisofpopulationstructures,inbreeding,and
familiallinkage
P152 FastlinkageanalysiswithMODscoresusingalgebraiccalculation
P153 Fetalexposuresandperinatalinfluencesontheprematureinfantmicrobiome
P154 Combininggenotypewithallelicassociationasinputforiterativepruningprincipal
componentanalysis(ipPCA)toresolvepopulationsubstructures
P155 Spuriouscrypticrelatednesscanbeinducedbypopulationsubstructure,population
admixtureandsequencingbatcheffects
P156 Effectofpopulationstratificationonvalidityofacase‐onlystudytodetectgene‐
environmentinteractions
P157 Anovelriskpredictionalgorithmwithapplicationtosmokingexperimentation
P158 Trio‐BasedWholeGenomeSequenceAnalysisofaCousinPairwithRefractoryAnorexia
Nervosa
P159 Powerandsamplesizeformulasfordetectinggeneticassociationinlongitudinaldatausing
generalizedestimatingequations
P160 Ontheevaluationofpredictivebiomarkerswithdichotomousendpoints:acomparisonof
thelinearandthelogisticprobabilitymodels
P161 Atwostagerandomforestprobabilitymachineapproachforepigenome‐wideassociation
studies
P162 Statisticalapproachesforgene‐basedanalysis:AcomprehensivecomparisonusingMonte‐
CarloSimulations
P163 ApolipoproteinEgenepolymorphismandleftventricularfailureinbeta‐thalassemia:A
meta‐analysis
P164 TheCooperativeHealthResearchinSouthTyrol(CHRIS)study
IndexbyCategories
IndexbyAuthors
Mini‐Symposium
Genomics‐basedPersonalizedMedicine
M1
Applicationofgenomictestsinbreastcancermanagement
MartinFilipits1
1MedicalUniversityofVienna,InstituteofCancerResearch,Vienna,Austria
Breastcancerisaheterogeneousdiseaseattheclinical,biologicalandparticularlyatthemolecular
level.Geneexpressionprofilinghasimprovedtheknowledgeonthecomplexmolecularbackground
ofthisdiseaseandallowsamoreaccurateprognosticationandpatientstratificationfortherapy.
Severalgenomictestshavebeendevelopedwiththeaimofimprovingprognosticinformation
beyondthatprovidedbyclassicclinicopathologicparameters.Someofthesetestsarecurrently
availableintheclinicandareusedtodetermineprognosisandmoreimportantlytoassistin
determiningtheoptimaltreatmentinpatientswithhormonereceptor‐positivebreastcancer.
Availabledatasuggestthatinformationgeneratedfromgenomictestshasresultedinachangein
decisionmakinginapproximately25%‐30%ofcases.Theclinicalrelevanceofgenomictestsand
theirabilitytodefineprognosisanddeterminetreatmentbenefitwillbediscussed.
M2
Riskpredictionmodelsusingfamilyandgenomicdata
JoanEBailey‐Wilson1
1ComputationalandStatisticalGenomicsBranch,NationalHumanGenomeResearchInstitute,National
InstitutesofHealth,Baltimore,UnitedStatesofAmerica
ComputationalandStatisticalGenomicsBranch,NationalHumanGenomeResearchInstitute,
NationalInstitutesofHealth,Baltimore,UnitedStatesofAmerica}{Advancesinourabilitytomodel
personalriskofdevelopingadiseasehaveacceleratedaslargeepidemiologicandgenomicstudies
haveincreasedourunderstandingofdiseasecausation.Predictionofdiseaseriskcanbebasedon
personalhistoryofenvironmentalexposures,familyhistoryofdiseaseandpersonalgenotypesat
geneticsusceptibilityloci.Approachestopredictingriskofdiseasethatutilizefamilialandgenetic
informationwillbediscussedforarangeofdifferentcausalmodelsfromsimpleMendelian
disordersthatarecausedbyvariantsinasinglegenetodiseasescausedbycomplexactionsof
multipleriskfactors.Theutilityofaddingfamilyhistoryandpersonalgenotypesintodiseaserisk
modelswillbecovered.Accuratediseaseriskpredictioncanbeimportanttoindividualhealthsince
itcanencourageindividualstohavemorefrequentscreeningprocedures,toundertake
environmentalriskreduction,andtoundergopreventivemedicalproceduresandtreatments.
M3
Theimportanceofappropriatequalitycontrolin‐omicsstudiesas
requiredforpersonalizedandstratifiedmedicine
BertramMüller‐Myhsok1,2,3
1MaxPlanckInstituteofPsychiatry,Munich,Germany
2MunichClusterforSystemsNeurology(SyNergy),Munich,Germany
3InstituteforTranslationalMedicine,UniversityofLiverpool,Liverpool,UnitedKingdom
ComputationalandStatisticalGenomicsBranch,NationalHumanGenomeResearchInstitute,
NationalInstitutesofHealth,Baltimore,UnitedStatesofAmerica}{Advancesinourabilitytomodel
personalriskofdevelopingadiseasehaveacceleratedaslargeepidemiologicandgenomicstudies
haveincreasedourunderstandingofdiseasecausation.Predictionofdiseaseriskcanbebasedon
personalhistoryofenvironmentalexposures,familyhistoryofdiseaseandpersonalgenotypesat
geneticsusceptibilityloci.Approachestopredictingriskofdiseasethatutilizefamilialandgenetic
informationwillbediscussedforarangeofdifferentcausalmodelsfromsimpleMendelian
disordersthatarecausedbyvariantsinasinglegenetodiseasescausedbycomplexactionsof
multipleriskfactors.Theutilityofaddingfamilyhistoryandpersonalgenotypesintodiseaserisk
modelswillbecovered.Accuratediseaseriskpredictioncanbeimportanttoindividualhealthsince
itcanencourageindividualstohavemorefrequentscreeningprocedures,toundertake
environmentalriskreduction,andtoundergopreventivemedicalproceduresandtreatments.
M4
StudyDesignsforPredictiveBiomarkers
AndreasZiegler1,2
1InstituteofMedicalBiometryandStatistics,UniversityofLübeck,UniversityMedicalCenterSchleswig‐
Holstein,CampusLübeck,Lübeck,Germany
2CenterforClinicalTrials,UniversityofLübeck,Lübeck,Germany
Biomarkersareofincreasingimportanceforpersonalizedmedicine,includingdiagnosis,prognosis
andtargetedtherapyofapatient.Examplesareprovidedforcurrentuseofbiomarkersin
applications.Itisshownthattheiruseisextremelydiverse,anditvariesfrompharmacodynamicsto
treatmentmonitoring.Theparticularfeaturesofbiomarkersarediscussed.Beforebiomarkersare
usedinclinicalroutine,severalphasesofresearchneedtobesuccessfullypassed,andimportant
aspectsofthesephasesareconsidered.Somebiomarkersareintendedtopredictthelikelyresponse
ofapatienttoatreatmentintermsofefficacyand/orsafety,andthesebiomarkersaretermed
predictivebiomarkersor,moregenerally,companiondiagnostictests.Usingexamplesfromthe
literature,differentclinicaltrialdesignsareintroducedforthesebiomarkers,andtheirprosand
consarediscussedindetail.
EducationalWorkshop
Pharmacogenomics:WhenDrugResponseGets
Personal
E1
Pharmacogenomics:Past,PresentandFuture
BrookeFridley1
1UniversityofKansasMedicalCenter,USA
Pharmacogeneticsisthestudyoftheroleofinheritanceinindividualgeneticvariationinresponse
todrugs.Inthispost‐genomicera,pharmacogeneticshasevolvedintopharmacogenomics,thestudy
oftheinfluenceofgeneticvariationacrosstheentiregenomeondrug‫ײ‬esponse.Pharmacogenomics
hasbeenheraldedasoneofthefirstmajorclinicalapplicationsofthestrikingadvancesthathave
occurredandcontinuetooccurinhumangenomicscience.Inthistalk,Iwillprovideanoverviewof
pharmacogenomicsanddiscussthepast,presentandfutureofpharmacogenomicsinthe21st
century.
E2
Assessingthegeneticbasisofdrugresponse
JohnWitte1
1UniversityofCalifornia,SanFrancisco,USA
Bydefinitionpharmacogenomictraitshaveanunderlyinggeneticbasis.Nevertheless,accurately
estimatingtheheritabilityofdrugresponseisimportantfordesigningstudiesandknowinghow
muchgeneticvariationcan‫د‬rhas‫ﺁ‬eenexplained.Unlikemostquantitativeandqualitativetraits,
however,responsetotreatmenthastwounique,complicatingfactors:itisagene‐druginteraction
andtheoutcomeisoftenintermsoftime‐to‐event.HereIwillpresentandapplymethodsthat
addressthesetwoaspectswhenestimatingthegeneticbasis(orheritability)ofpharmacogenomic
traits.
E3
ClinicalUtilityinPharmacogenomics:GettingBeyondIndividualVariants
HaeKyungIm1
1UniversityofChicago,USA
Studiesinpharmacogenomicshaveidentifiedmanyindividualvariantswithsufficientlylargeeffect
sizestohaveclinicalutility,andmanyofthesearenowthesubjectofimplementationstudiesata
varietyoflevels.Recentresearchoncommondiseasesandcomplextraitshave,however,raisedthe
possibilitythatmixedmodelsallowingseparatelyforthecontributionofvariantswithlargereffect
sizesandapolygenicbackgroundmayyieldimprovedprediction.Aswemedicalcentersroutinely
movetohavinglarge‐scalegenomedataroutinelyavailableonpatients,asopposedtoone‐off
genotypingfortheprescribingofspecificdrugs,theopportunitytobuildpredictorsofadverse
eventsandefficacyusinglargescalegenomedataratherthanindividual(orsmallnumbersof)
variantsbecomesarealpossibility.Usingrealexamplesfromlarge‐scalestudies,wewillcontrast
predictionbasedonindividualorsmallnumbersofvariantswithpredictionsbasedonlarge‐scale
information.WewillalsodiscusseffortstoimplementthesealternativeapproachesinEMRsettings.
E4
Smokingbehaviorandlungcancerriskrelatedtonicotinicacetylcholine
receptorvariantsandmetabolicvariants.
ChristopherIAmos1
1GeiselSchoolofMedicine,USA
InthispresentationIcontrastthediscoveryofgeneticvariantsthatinfluencesmokingbehavior
includinginitiation,dailyconsumptionandcessation.Themostprominentassociationsarewiththe
nicotinicacetylcholinereceptorgenefamilyonchromosome15q25.1.Thesegenesalongwith
CYP2A6stronglyinfluencesmokingbehaviorandalsoaffectlungcancerrisk.Iwilldescribethe
strikingimpactthatvariationinthesegenesappearstohaveontheefficacyofpharmacological
interventionstoinfluencesmokingcessation.Finally,Iwilldescribestudiesoflungcancerriskand
howthesegenesrelatetoit,alongwithafurtherdiscussionofthepotentialrelevanceofnovel
associationsrecentlydiscoveredforsquamouslungcancerthatmayinfluencechemotherapeutic
responses.
InvitedSpeakers
I1
Enrichmentdesignsforthedevelopmentofpersonalizedmedicine
MartinPosch1
1ViennaMedicalUniversity,Austria
Iftheresponsetotreatmentdependsongeneticbiomarkers,itisimportanttoidentify
(sub)populationswherethetreatmenthasapositivebenefitriskbalance.Oneapproachtoidentify
relevantsubpopulationsaresubgroupanalyseswherethetreatmenteffectisestimatedin
biomarkerpositiveandbiomarkernegativegroups.Subgroupanalysisarechallengingbecause
differenttypesofrisksareassociatedwithinferenceonsubgroups:Ontheonehand,ignoringa
relevantsubpopulationonecouldmissatreatmentoptionduetoadilutionofthetreatmenteffectin
thefullpopulation.Even,ifthedilutedtreatmenteffectcanbedemonstratedinanOverall
population,itisnotethicaltotreatpatientsthatdonotbenefitfromthetreatmemt,iftheycanbe
identifiedinadvance.Ontheotherhandselectingaspurioussub‐populationisnotwithoutrisk
either:itmightincreasetherisktoapproveaninefficienttreatment(inflatingthetype1errorrate),
ormaywronglyleadtorestrictinganefficienttreatmenttoatoonarrowfractionofapotential
benefitingpopulation.Thelattercannotonlyleadtoreducedrevenuefromthedrug,butisalso
unfavourablefromapublichealthperspective.Weinvestigatetheserisksfornon‐adaptivestudy
designsthatallowforinferenceonsubgroupsusingmultipletestingproceduresaswellasadaptive
designs,wheresubgroupsmaybeselectedinaninterimanalysis.Quantifyingtheriskswithutility
functionsthecharacteristicsofsuchadaptiveandnon‐adaptivedesignsarecomparedforarangeof
scenarios.
I2
Causalassociationstructuresin‐omicsdata:howfarcanwegetwith
statisticalmodeling?
KristaFischer1
1TartuUniversity,Estonia
Thistalkmainlyconcentratesonthesettingwhereassociationofonegenotypemarker(typically
SNP)withtwocorrelatedphenotypesisstudied.Inso‐called"MendelianRandomization"studies
themainparameterofinterestcorrespondstoacausaleffectofonephenotypictraitonanother
trait,whereasageneticmarkerisusedasaninstrument.Despiteoftheincreasingnumberof
publicationsusingthismethodologicalapproach,theunderlyingassumptionsareoftenoverlooked.
Therefore,manyofthepublishedeffectestimatesmayactuallybebiasedandmisleading.Oneofthe
mainuntestableassumptionsisthe"nopleiotropy"assumption‐thegenotypehasadirectcausal
effectononephenotypeonly,whereastheeffectonthesecondphenotypeisfullymediatedbythe
firstone.
Whenthisisnotfulfilled,thegenotypeissaidtohaveapleiotropiceffectonbothphenotypes,
whereasanotherclassofmodelsisbeendesignedtoestimatesucheffects.However,wewillshow
thatmathematicallyonecannotdistinguishbetweenthetwomodels:themodelunderlyingthe
MendelianRandomizationscenarioandthemodelforpleiotropiceffect.Wewilldiscusswhether
somesensitivityanalysismethodsmayhelptodrawacorrectconclusionhere.
Inaddition,wediscussanotherassumptionunderlyingtheMendelianRandomizationidea:the"no‐
treatment‐effectheterogeneity"assumption.Hereaparallelcanbedrawnwithrandomizedclinical
trials,wherethisassumptioniscrucialtoallowforactivetreatmentonthecontrolarm.Usingalso
simulationresults,theeffectofdeviationsfromthisassumptionisstudied.
I3
Therelevanceofepigenomicsforpersonalizedmedicine
ChristophBock1
1ResearchCenterforMolecularMedicineoftheAustrianAcademyofSciences,Austria
Inmypresentation,Iwillsummarizetheroleofnextgenerationsequencingforpersonalized
medicineandhighlighttherelevanceofbioinformaticandbiostatisticalmethodsforinterpreting
thevastamountofgenome,epigenomeandtranscriptomedatathatarebeinggeneratedatCeMM
andatmanygenomicsinstitutesworld‐wide.Thetalkwillalsodiscussourongoingworkwiththe
EuropeanBLUEPRINTprojectconsortium(http://blueprint‐epigenome.eu/)aimedatestablishing
comprehensiveepigenomemapsofhematopoieticcelltypesandvarioustypesofleukemiacells.I
willconcludebyoutlininganintegratedcomputational/experimentalapproachtowardrational
designofepigeneticcombinationtherapies(BockandLengauer2012NatureReviewsCancer),
whichwepursueincollaborationbetweentheCeMMResearchCenterforMolecularMedicineand
theMedicalUniversityofVienna.
I4
Finemappingofcomplextraitlociwithcoalescentmethodsinlargecase‐
controlstudies
ZiqianGeng1,PaulScheet1,SebastianZöllner1
1UniversityofMichigan,USA
Case‐controlstudiesarewidelyusedtoidentifygenomicregionscontainingdiseasevariants.
However,identifyingtheunderlyingriskvariantsforcomplexdiseasesischallengingduetothe
complicatedgeneticdependencestructurecausedbylinkagedisequilibrium(LD).Bymodelingthe
evolutionaryprocessofatargetregion,coalescent‐basedapproachesimprovethisidentificationby
usingallavailablehaplotypeinformation.Suchmethodsestimatethegenealogyatallsitesinthe
regionandthusmodeltheprobabilityofcarryingriskvariantsatalllocijointly.Fromthese
probabilitiesweobtainBayesianconfidenceintervals(CIs)wheretrueriskvariantsaremostlikely
tooccur.Additionally,thegenealogyateachpositionprovidesmoreinformationabouttheshared
ancestryofneighboringsites.Indeed,suchcarefulmodelingofthesharedancestryofsequencesis
alsobeneficialinhaplotypingandvariantcallinginregionsofinterests(ROI)wheretraditional
hiddenMarkovapproachesstruggle.However,existingcoalescent‐basedmethodsare
computationallyverychallengingandcanonlybeappliedtosamplesbelow200individuals.Here,
weproposeanovelapproachtoovercomethisdifficulty,sothatitcanbeappliedtolarge‐scale
studies.First,weinferasetofclustersfromthesampledhaplotypessothathaplotypeswithineach
clusterareinheritedfromacommonancestor.Then,weapplycoalescent‐basedapproachesto
approximatethegenealogyofancienthaplotypesatdifferentpositionsacrosstheROI.Doingso,the
dimensionofexternalnodesincoalescentmodelsisreducedfromthetotalsamplesizetothe
numberofclusters.Finally,weevaluatetheposition‐specificclustergenealogyandtheir
descendants’phenotypedistribution,tointegrateoverallpositionsandestablishCIswhererisk
variantsaremostlikelytooccur.Insimulationstudies,ourmethodcorrectlylocalizesshort
segmentsaroundtrueriskpositionsforbothrare(1%)andcommon(5%)riskvariantsindatasets
withthousandsofindividuals.Insummary,wehavedevelopedanovelapproachtoestimatethe
genealogythroughoutsequencedregions.Infinemappingofcomplextraitloci,ourmethodis
applicableforlarge‐scalecase‐controlstudiesusingsequencingdata.
I5
Theinterfacehypothesisinexplaininghost‐bacterialinteractionsinthe
humangut
KnutRudi1
1NorwegianUniversityforLifeSciences,Norway
Ourgutmicrobiotaistremendouslycomplex,outnumberingthehostcellsbyafactoroftenandthe
numberofgenesbyafactorofonehundred.Thegutmicrobiotaservesthemainfunctionsof
extractingenergyfromthefood,productionofvitaminsandother(essential)biomolecules,in
additiontoprotectiontowardspathogens.However,despitemajoreffortswedostillnotknowthe
basicmechanismsforhost‐bacterialinteractionsinthegut.Wehavethereforerecentlyproposed
theinterfacehypothesis,advocatingtheimportanceofpositivehostselectionformutualisticgut
bacteria.Iwillpresentdetailsaboutthehypothesis,andhowitissupportedfromthecurrent
knowledgeaboutthehumangutmicrobiota.
NeelandWilliamsAwardCandidates
A1
Anovelmethodusingcrosspedigreesharedancestrytomaprarecausal
variantsinthepresenceoflocusheterogeneity
HaleyJAbel1,MichaelAProvince1
1DivisionofStatisticalGenomics,WashingonUniversitySchoolofMedicine,St.Louis,MO
Currently,thereisgreatinterestintheuseoffamilystudiestoidentifyrarevariantsunderlying
complexdisease.However,attemptsatfinemappingareconfoundedbylocusheterogeneity,which
resultsinnoisyandpoorlylocalizedlinkagesignals,andanabundanceofrarevariants,which
frequentlysegregatewithphenotypebychance.Asrarevariantssharedacrosspedigreesarelikely
tobeofrecentorigin,wehavedevelopedanapproachleveragingidentity‐by‐descent(IBD)
betweenpedigreefounderstobetterlocalizelinkagesignalsinthepresenceofheterogeneity.Our
methodreliesonsegmentssharedidentically‐by‐state(IBS)acrosspedigrees:itoptimizesover
pedigreemembersandcalculatesascorebasedonthesumofmaximalpairwisesharedlengthsat
eachlocus.UseofunphasedIBSmakesitbothcomputationallyefficient,sothatpedigree‐based
permutationtestsassessingsignificancearetractable,androbusttogenotypingandhaplotype
phaseswitcherrors.Moreover,ourmethodprovidesacross‐familymetrictopermitlocalclustering
offamiliesnearIBDregions:thisallowsstratificationbyrecentsharedancestry,and,insimulations,
accuratelyrecoversancestralrelationships.Wehaveevaluatedtheperformanceofourmethodby
coalescentsimulationoffounderindividuals,followedbygene‐droppingontopedigrees.Undera
varietyofscenarios,withrarecausalvariants(MAF<0.01)andmodesteffectsizes(OR=5‐7),our
approachachieves60‐80%power,andisabletodetectsharedancestralsegmentsharboringrare
causalvariantswheremultipointlinkageandrare‐variantburdentestsfail.
Categories: CoalescentTheory,Heterogeneity,Homogeneity,LinkageandAssociation
A2
Survivalanalysiswithdelayedentryinselectedfamilieswithapplication
tohumanlongevity.
MarRodriguezGirondo1,JeanineHouwing‐Duistermaat1
1LUMC,TheNetherlands
Althoughthereisevidencefromseveralstudiesthatlongevityaggregateswithinfamilies,
identificationofgeneticfactorshasnotbeensuccessful.Reasonsforlackofprogressmightbethead
hocdefinitionofbeingolderthanaspecificthreshold(e.g.olderthan90yearsofage).Asalternative
wewillconsidersurvivalmodelsfortheanalysisoflongevityinfamilystudies.Challengesareto
modeltheascertainmentofthefamilies,totakeintoaccountcorrelationbetweenfamilymembers
andtodealwithdelayedentry.Methodsforsurvivalanalysiswithdelayedentryinsmallclusters
areavailable(e.g.Rondeauetal,2012).Thesemethodsprovidebiasestimatesforlargerclusters
(Jensenetal,2004),becausetheydonotadjustforascertainment.WeproposeaCoxmodelwitha
frailtyandwithinverseprobabilityweightingtoaccountfortheselectionofthefamiliesandthe
delayedentry.Theweightswillbebasedonthelatentfrailtiesinaproportionalhazardsmodel.Via
simulationsweshowedthatourapproachperformsbetter(lessbias)thanexistingmethodsfor
largefamilies(>8subjects)andlargefrailties(>0.5).ThisworkismotivatedbytheLeiden
Longevitystudycomprising420familieswithatleasttwononagenariansiblings.Thesizeof
sibshipswithmemberswhobecomeolderthan60yearsvariesfrom2to13siblings.Themaximum
observedageis107yearsand13%ofthenonagenariansisstillalive.Weestimatedtheeffectof
APOEE4alleleonsurvival.Theestimateofthevarianceofthefrailtywas0.082and0.230forthe
standardapproachandourapproachrespectively.Theestimateoftheloghazardratiowas‐0.272
(s.e.0.113)and‐0.212(s.e.0.070)forthestandardandourapproachrespectively.
Categories: Ascertainment,Association:Family‐based,Heritability
A3
Combiningfamily‐andpopulation‐basedimputationdataforassociation
analysisofrareandcommonvariantsinlargepedigrees
MohamadSaad1,EllenMWijsman1
1DepartmentofBiostatistics,UniversityofWashington
Inthelasttwodecades,complextraitshavebecomethemainfocusofgeneticstudies.The
hypothesisthatbothrareandcommonvariantsareassociatedwithcomplextraitsisincreasingly
beingdiscussed.Family‐basedassociationstudiesusingrelativelylargepedigreesaresuitablefor
bothrareandcommonvariantidentification.Becauseofthehighcostofsequencingtechnologies,
imputationmethodsareimportantforincreasingtheamountofinformationatlowcost.Arecent
family‐basedimputationmethod,GIGI,isabletohandlelargepedigreesandaccuratelyimputerare
variants,butdoeslesswellforcommonvariantswherepopulation‐basedmethods(e.g.;BEAGLE)
performbetterandcanalsobeused.Weproposeaflexibleapproachtocombineimputationdata
fromfamily‐andpopulation‐basedmethods.Weselect,foreverySNPandeverysubject,thesetof3
genotypeposteriorprobabilitiesfromthemethodwiththehighestvarianceoftheseprobabilities.
WealsoextendtheassociationtestSKAT‐RC,originallyproposedfordatafromunrelatedsubjects,
tofamilydatawithcontinuoustraitinordertomakeuseofsuchimputeddata.Wecallthis
extension“famSKAT‐RC”.WecomparetheperformanceoffamSKAT‐RCandseveralotherexisting
burdenandkernelassociationtests.Insimulatedpedigreesequencedata,ourresultsshowan
increaseofimputationaccuracyfromthecombinedapproach.Also,thedatashowanincreaseof
poweroftheassociationtestswiththisapproachovertheuseofeitherfamily‐orpopulation‐based
imputationmethodsalone,inthecontextofrareandcommonvariantsinasinglegene.Moreover,
ourresultsshowedbetterperformanceoffamSKAT‐RCcomparedtotheotherconsideredtests,in
mostscenariosinvestigated.
Categories: Association:Family‐based,MultipleMarkerDisequilibriumAnalysis,SequencingData
A4
Mixedmodelingfortime‐to‐eventoutcomeswithlarge‐scalepopulation
cohortsandgenome‐widedata
ChristianBenner1,2,MattiPirinen1,EmmiTikkanen1,2,SamuliRipatti1,2,3
1InstituteforMolecularMedicineFinland(FIMM),UniversityofHelsinki,Helsinki,Finland
2HjeltInstitute,UniversityofHelsinki,Helsinki,Finland
3WellcomeTrustSangerInstitute,Hinxton,Cambridge,UK
Recentdevelopmentonlinearmixedmodelshasprovidedacommonframeworkforheritability
estimation,multi‐locusassociationtestingandgenomicpredictionofquantitativetraitsin
populationcohortsofunrelatedindividuals.Thepossibilitytousepopulationcohortsratherthan
familystructurescouldopenupnewavenuesalsofortime‐to‐eventoutcomesingenetic
epidemiology.However,connectingtime‐to‐eventoutcomestobiggenomicsdatahassofarnot
beencomputationallyfeasiblewithhithertoexistingsoftware.Weintroduceanovelsurvival
analysismethodforheritabilityestimation,multi‐locusassociationtestingandgenomicrisk
predictionthatscalestomillionsofgeneticmarkersandhealtheventsintensofthousandsof
individuals.MotivationforourworkcomesfromalargeanduniquecollectionofFinnishpopulation
cohortsforwhichwehavebothdetailedgenomicandcomprehensivehealthregistrydata.Our
approachimplementsaveryflexiblepiecewiseconstanthazardmodelthatcontainsanindividual‐
specificGaussianrandomeffectwithanarbitrarycovariancestructure.Computationally,we
transformtheproblemtoaPoissonmodel,whichweanalyzebyfittingahierarchicalgeneralized
linearmodel.Wedemonstratetheruntimeefficiencyofourmethodandgiveanexampleof
heritabilityestimationandmulti‐locusassociationtestingforcardiovasculardiseaserelatedevents
usingupto16,000Finnishindividuals.Ourworkextendsthecomputationaltractabilityoflinear
mixedmodelsfromquantitativetraitstotime‐to‐eventoutcomesandwillproveuseful,e.g.,for
combininginformationacrossindividuals’genomesandtheirhospitalrecords.
Categories: Association:Genome‐wide,CardiovascularDiseaseandHypertension,Heritability,
MaximumLikelihoodMethods
A5
Thecollapsedhaplotypepatternmethodforlinkageanalysisofnext‐
generationsequencingdata
GaoTWang1,DiZhang1,BiaoLi1,HangDai1,SuzanneMLeal1
1BaylorCollegeofMedicine
Traditionally,linkageanalysiswasusedtomapMendeliandiseases.Nextgenerationsequencing
(NGS)makesitpossibletodirectlysequenceindividualswithMendeliandiseasesandidentify
causalvariantsbyfiltering.LinkageanalysisofSNPdataaresometimesusedinconjunctionwith
NGStoincreasethesuccessofidentifyingthecausalvariant.WiththereductionincostofNGS,DNA
samplesfrommultiplefamiliescanbesequencedandlinkageanalysiscanbeperformeddirectly
usingNGSdata.Inspiredby“burden”testsforcomplextraitrarevariantassociationstudies,we
developedthecollapsedhaplotypepattern(CHP)methodtogeneratemarkersfromsequencedata
forlinkageanalysis.TodemonstratethepoweroftheCHPmethodweanalyzedandperformed
powercalculationsusingdatafromseveraldeafnessgenes.PoweranalysisshowedthattheCHP
methodissubstantiallymorepowerfulthananalyzingindividualSNVs.Specificallyforanautosomal
recessivemodelwithallelicheterogeneityandlocusheterogeneityof50%,itrequires12families
fortheCHPmethodtoachieveapowerof90%fortheSLC26A4gene,whileanalyzingindividual
SNVsrequires>50familiestoachievethesamepower.Unlikethecommonlypracticedfiltering
approachesusedforNGSdata,theCHPmethodprovidesstatisticalevidenceoftheinvolvementofa
geneinMendeliandiseaseetiology.Additionallybecauseitincorporatesinheritanceinformation
andpenetrancemodelsitislesslikelythanfilteringtoexcludecausalvariantsinthepresentsof
phenocopiesand/orreducedpenetrance.WerecommendtheuseoftheCHPmethodinparallelto
filteringmethodstotakefulladvantageofthepowerofNGSinfamilies.
Categories: LinkageAnalysis,SequencingData
A6
Meta‐analysisapproachforhaplotypeassociationtests:ageneral
frameworkforfamilyandunrelatedsamples
ShuaiWang1,JingHZhao2,MarkOGoodarzi3,JoséeDupuis1
1DepartmentofBiostatistics,BostonUniversitySchoolofPublicHealth,Boston,MA
2MRCEpidemiologyUnit,UniversityofCambridge,InstituteofMetabolicScience,Addenbrooke'sHospital,Box
285HillsRoad,Cambridge,UnitedKingdom
3DivisionofEndocrinology,DiabetesandMetabolism,Cedars‐SinaiMedicalCenter,LosAngeles,CA
Meta‐analysishasbeenwidelyusedtoimprovepowertodetectassociatedvariantsingenome‐wide
associationstudies.Severalmeta‐analysismethodshavebeendevelopedandsuccessfullyappliedto
combineassociationtestsofsinglevariantandgene‐basedtestsfrommultiplecohorts.However,
meta‐analysisofhaplotypeassociationresultsremainsachallenge,becausedifferenthaplotypes
maybeobservedacrosscohorts.Weproposeatwo‐stagemeta‐analysisapproachtocombine
haplotypeanalysisresults.Ourapproachallowseachcohorttocontributeassociationresultsfrom
uniquelyobservedhaplotypes,inadditiontohaplotypesobservedinmultiplecohorts.Inthefirst
stage,eachcohortcomputestheexpectedhaplotypeeffectsinaregressionframework,selectingthe
mostfrequenthaplotype,whichcanvaryacrosscohorts,asthereferencehaplotypeandincludinga
randomfamilialeffecttoaccountforrelatedness,ifappropriate.Forthesecondstage,weproposea
multivariategeneralizedleastsquaremeta‐analysisapproachtocombinehaplotypeeffectsfrom
multiplecohorts.Associationtestsforeachhaplotypeandaglobaltestcanbeobtainedwithinour
framework.AsimulationstudyshowsthatourapproachhasthecorrecttypeIerror.Wepresentan
applicationtogenotypesfromIlluminaHumanExomeBeadchiparray,whereweassessthe
associationbetweenhaplotypesformedbyrarevariantsinafastingglucose‐associatedlocus
(G6PC2).Wethencombinedhaplotypeanalysisresultsfrom18CHARGE(CohortsforHeartand
AgingResearchinGenomicEpidemiologyConsortium)cohorts.Theglobalhaplotypeassociation
testishighlysignificant(p=1.1e‐17),andmoresignificantthananysingle‐variantandgene‐based
tests.
Categories: Association:Family‐based,Diabetes,HaplotypeAnalysis,QuantitativeTraitAnalysis
ContributedPlatformPresentations
C1
Identificationofbloodpressure(BP)relatedcandidategenesby
population‐basedtranscriptomeanalyseswithintheMetaXpress
Consortium
ChristianMüller1,KatharinaSchramm2,ClaudiaSchurmann3,SoonilKwon4,ArneSchillert5,
ChristianHerder6,GeorgHomuth3,SimoneWahl7,HaraldGrallert7,AndreasZiegler5
1GeneralandInterventionalCardiology,UniversityHeartCenterHamburg,Germany
2InstituteofHumanGenetics,HelmholtzCenterMunich,GermanResearchCenterforEnvironmentalHealth,
Neuherberg,Germany
3InterfacultyInstituteforGeneticsandFunctionalGenomics,UniversityMedicineandErnst‐Moritz‐Arndt‐
UniversityGreifswald,Greifswald,Germany
4MedicalGeneticsInstitute,Cedars‐SinaiMedicalCenter,LosAngeles,USA
5InstituteofMedicalBiometryandStatistics,UniversityofLübeck,UniversityMedicalCenterSchleswig‐
6InstituteforClinicalDiabetology,GermanDiabetesCenter,LeibnizCenterforDiabetesResearchatHeinrich
HeineUniversityDüsseldorf,Germany
7ResearchUnitofMolecularEpidemiology,HelmholtzZentrumMünchen,GermanResearchCenterfor
EnvironmentalHealth,Neuherberg,Germany
Highbloodpressure(BP)isaglobalmajorriskfactorforcardiovasculardiseases.Weanalyzed
associationsbetweenthetranscriptomeandBPtraitsinlargecohortsoftheMetaXpress
Consortium.TranscriptomicdatafromtheIlluminaHumanHT‐12BeadChiparraywereavailablefor
4533individualsfromthethreeGermancohortsandoneUScohort.Expressionlevelswere
measuredinmonocyte(n=2549,GutenbergHealthStudy(GHS))andwholebloodcell(n=1984
CooperativeHealthResearchintheRegionofAugsburg(KORAF4)andStudyofHealthin
Pomerania(SHIP‐TREND),Multi‐EthnicStudyofAtherosclerosis(MESA)).Associationstosystolic
BP(SBP),diastolicBP(DBP)andpulsepressure(PP)wereestimatedbylinearregressionwith
adjustmentsforsex,age,bodymassindex(BMI),RNAstoragetime,amplificationlayoutandRNA
integritynumberwithineachstudy.ApooledanalysiswasconductedwithinGHSandMESAusing
theinversevariancemethod.Significantassociations(FDR≤0.05)wereselectedforreplicationin
KORAF4andSHIP‐TREND.Geneswithconsistenteffectdirectionsandp≤0.05inbothinitialstudies
wereselectedascandidates.Intotal,8uniquegeneswereconsistentlyassociatedwithsystolic
bloodpressure(SBP),diastolicbloodpressure(DBP)orpulsepressure(PP)inbothdiscoveryand
replicationsteps:CEBPA,CRIP1,F12,LMNA,MYADM,TIPARP,TPPP3andTSC22D3.Intotal,the
candidategenesexplainedbetween4‐13%,4‐6%and2‐8%ofinter‐individualvarianceofSBP,DBP
andPP,respectively.ThisisthefirststudyinvestigatingtheassociationsbetweenBPtraitsand
wholetranscriptomesinmorethan4000individuals.Thecomprehensiveanalyseshighlighteight
geneswhichareassociatedwithBP.
Categories: Association:CandidateGenes,Association:Genome‐wide,CardiovascularDiseaseand
Hypertension,GeneExpressionArrays,GeneExpressionPatterns
C2
Mixed‐modelanalysisofcommonvariationrevealspathwaysexplaining
varianceinAMDrisk
JacobBHall1,MargaretAPericak‐Vance2,WilliamKScott2,JaclynLKovach2,StephenDSchwartz2,
AnitaAgarwal3,MilamABrantley3,JonathanLHaines1,WilliamSBush1
1InstituteforComputationalBiology,CaseWesternReserveUniversity,Cleveland,OH
2JohnPHussmanInstituteforHumanGenomics,UniversityofMiamiMillerSchoolofMedicine,Miami,FL
3DepartmentofOphthalmologyandVisualSciences,VanderbiltUniversity,Nashville,TN
Age‐relatedmaculardegeneration(AMD)istheleadingcauseofirreversibleblindnessintheelderly
indevelopedcountriesandcanaffectmorethan10%ofindividualsoverage80.AMDhasalarge
geneticcomponent,withheritabilityestimatedtobebetween45%&70%.Numerouslocihave
beenidentifiedandimplicatevariousmolecularmechanismsandpathwaysinAMDpathogenesis.
Eightpathways,includingangiogenesis,antioxidantactivity,apoptosis,complementactivation,
inflammatoryresponse,nicotinemetabolism,oxidativephosphorylation,andthetricarboxylicacid
cycle,wereselectedforourstudybasedonanextensiveliteraturereview.Whilethesepathways
havebeenproposedinliterature,theoverallextentofthecontributiontoAMDheritabilityforeach
pathwayisunknown.Inacase‐controldataset,weusedGenome‐wideComplexTraitAnalysis
(GCTA)toestimatetheproportionofvarianceinAMDriskexplainedbyallSNPsineachpathway.
SNPswithina50kbregionflankingeachgenewereassessed,aswellasmoredistant,putatively
regulatorySNPs,basedondatafromtheENCODEproject.Wefoundthat19establishedAMDrisk
SNPscontributedto13.3%ofthevariationinriskinourdataset,whiletheremaining659,181SNPs
contributedto36.7%.Adjustingforthese19riskSNPs,thecomplementactivationand
inflammatoryresponsepathwaysexplainedastatisticallysignificantproportionofadditional
varianceinAMDrisk(9.8%and17.9%,respectively),withotherpathwaysshowingnosignificant
effects(0.3%–4.4%).Ourresultsshowthatadditionalvariantsassociatedwithcomplement
activationandinflammationgenescontributetoAMDrisk,andthatthesevariantsarelikelyin
codingandnearbyregulatoryregions.
Categories: Case‐ControlStudies,Heritability,MaximumLikelihoodMethods,MultilocusAnalysis,
Pathways
C3
APhenome‐WideAssociationStudyofNumerousLaboratoryPhenotypes
inAIDSClinicalTrialsGroup(ACTG)Protocols
AnuragVerma1,SarahAPendergrass2,EricSDaar3,RoyMGulick4,RichardHaubrich5,GregoryK
Robbins6,DavidWHass7,MarylynDRitchie1
1ThePennsylvaniaStateUniversity,UniversityPark,Pennsylvaina,USA
2ThePennsylvaniaStateUniversity,UniversityPark,PA,USA
3DepartmentofMedicine,LosAngelesBiomedicalResearchInstitute,Harbor‐UCLAMedicalCenter,Torrance,
California,USA
4WeillMedicalCollegeofCornellUniversityNewYork,NewYork,USA
5UniversityofCaliforniaSanDiego,SanDiego,California,USA
6DepartmentofMedicine,MassachusettsGeneralHospital,HarvardMedicalSchool,Boston,Massachusetts,
USA
7VanderbiltUniversity,Nashville,Tennessee,USA
Phenome‐WideAssociationStudies(PheWAS)havethepotentialtoefficientlydiscovernovel
geneticassociationsacrossmultiplephenotypes.Prospectiveclinicaltrialsdataofferaunique
opportunitytoapplyPheWAStopharmacogenomics.HerewedescribethefirstPheWAStoexplore
associationsbetweengenotypicdataandclinicaltrialdata,bothpre‐treatmentandfollowing
initiationofantiretroviraltherapy.A"pre‐treatment"PheWASconsidered27laboratoryvariables
from2807subjectswhohadparticipatedin4ACTGprotocols(ACTG384,A5142,A5095and
A5202),andanalyzed~5MimputedSNPs.Lowestp‐valueswereforpre‐treatmentbilirubin,
neutrophilcounts,andHDLcholesterollevels.Theseandmultipleotherlaboratoryvariables
matchedassociationsintheNHGRIGWASCatalog.An"on‐treatment"PheWASconsidereddatafrom
1181subjectsfromA5202.Weconsidered838phenotypesandsub‐phenotypesderivedfrom6
variables:CD4counts,HIVcontrol,fastingLDL,fastingtriglycerides,efavirenzpharmacokinetics
(PK),andatazanavirPK.Weconsidered2,374annotateddrug‐relatedSNPsfromPharmGKB.Of23
associationswiththelowestp‐values(byphenotype),21(91%)werewithgeneswithmatching
biologicalplausibility:LDLwithLPLandAPOE;triglycerideswithLPL;CD4countswithinnate
immuneresponsegeneTNF,HIVcontrolwithadaptiveimmuneresponsegeneHLA‐DRQA1,
efavirenzPKwithCYP2B6;atazanavirPKwithdrugtransportergeneABCC4.Thisanalysis
highlightsthepotentialutilityofPheWAStoevaluateclinicaltrialsdatasetsforgeneticassociations.
Categories: Association:CandidateGenes,Association:Genome‐wide,Association:UnrelatedCases‐
Controls,Bioinformatics,Case‐ControlStudies,Epigenetics,MultivariatePhenotypes,Population
Genetics,PopulationStratification
C4
eMERGEPhenome‐WideAssociationStudy(PheWAS)IdentifiesClinical
AssociationsandPleiotropyforFunctionalVariants
AnuragVerma1,ShefaliSVerma1,SarahAPendergrass1,DanaCCrawford2,DavidRCrosslin3,
HelenaKuivaniemi4,WilliamSBush2,YukiBradford5,IftikharKullo6,SueBielinski6
1CenterforSystemsGenomics,DepartmentofBiochemistryandMolecularBiology,PennsylvaniaState
University,UniversityPark,PA,USA
2CaseWesternUniversity,Cleveland,OH,USA
3DepartmentofMedicine,DivisionofMedicalGenetics,UniversityofWashington,Seattle,WA,USA
4GeisingerHealthSystem,Danville,PA,USA
5VanderbiltUniversity,Nashville,TN
6MayoClinic,Rochester,MN,USA
Weperformedaphenome‐wideassociationstudy(PheWAS)exploringtheassociationbetween
stop‐gainedgeneticvariantsandacomprehensivegroupofphenotypestoidentifynovel
associationsandpotentialpleiotropy.Usingmultiplebioinformaticstoolsweselected38
functionallyrelevantstop‐gained/nullgeneticvariantswithinthegenotypicdataof37,972
unrelatedpatientsfromsevenstudysitesintheElectronicMedicalRecordsandGenomics
(eMERGE)Network.Wecalculatedcomprehensiveassociationsbetweenthesevariantsandcase‐
controlstatusfor3,518ICD9diagnosiscodes(requiring≥3visitsperindividualtoidentifycase
status,≥10casesubjectsperICD9code).Associationswereadjustedforage,sex,site,platformand
thefirst3principalcomponents.Atotalof418associationspassedaliberalsignificancethreshold
ofp<0.01.ThemostsignificantassociationwasbetweenGLG1rs9445and“chronicnon‐alcoholic
liverdisease”(p=4.12x10‐5,β=2.60).Weidentifiedmanypotentiallypleiotropicassociationsatp<
0.01,35outof38SNPsdemonstratedassociationswithmorethanonephenotype,and17SNPs
wereeachassociatedwith>10differentICD9codes.Forexample,wefoundassociationsforIL34
rs4985556with25diagnoses,suchas“lupuserythematosus”(p=5.94x10‐3,β=0.98)andforGBE1
rs2229519with33diagnoses,suchas“hypertension”(p=1.2x10‐3,β=0.067),“hyperlipidemia”
(p=6.66x10‐3,β=0.058),and“ocularhypertension”(2.49x10‐3,β=0.21).Wewillseekreplicationof
theseresults.Inconclusion,ourPheWASshowsstop‐gainedvariantsmayhaveimportant
pleiotropiceffects,andthatPheWASareapowerfulstrategytominethefullpotentialoftheEMR
forgenome‐phenomeassociations.
Categories: Association:Genome‐wide,MultilocusAnalysis
C5
AnovelG‐BLUP‐likephenotypepredictorleveragingregionalgenetic
similarityanditsapplicationsinpredictingdiseaseseverityanddrug
response
QuanLong1,EliAStahl2,JunZhu1
1DepartmentofGeneticsandGenomicSciences,IcahnSchoolofMedicineatMountSinai
2DepartmentofPsychiatry,IcahnSchoolofMedicineatMountSinai
Clinicalusesofphenotypepredictionsbasedongenotype(e.g.,PersonalizedMedicine)are
emerging,empoweredbyhigh‐throughputtechnology.Itiswellknownthatdiseaseseverityand
drugresponsediffersignificantlyacrosspopulationsorindividualpatientswithdifferentgenetic
background.Predictingsuchphenotypesusingseeminglyunrelatedsamples,thenstratifying
patientsbasedonthesepredictions,couldbecrucialforthedesignofclinictrails.Therearetwo
majoractivebranchesofgenotype‐basedphenotypepredictionsbasedonwholegenome
regression.Oneismodelselection,inwhichallgeneticmarkersaremodeled(usuallyinconjunction
withBayesianorothervariableselectioncriteria),andwhichmaysufferfromoverfittingdueto
astronomicalnumberofcombinationsofvariables/markers;theotherisG‐BLUPbasedonrandom
effectsregression,fittingphenotypicvariancebykinshipmatrixofthesampleestimatedfrom
genotypicsimilarity,whichmayruntheriskofunderfittingforcomplextraitsforwhich
infinitesimalmodeldoesnothold.WedevelopedaG‐BLUPlikepredictorthatstrikesthebalanceon
theabovetrade‐off.BasedonGWASsignalsorbiologicalaprioriknowledge,afewregionsare
selectedandtheirphenotypiccontributionsestimatedbyG‐BLUP.Then,modelselectionisapplied
tospecifyweightsforthedifferentregions.Usingsimulations,wedemonstratethatthepresent
predictorsignificantlyimprovespredictionpoweringeneralandinvestigateconditionsunder
whichitperformsbestornotcomparedwithpuremodelselectionorstandardG‐BLUP.Weapply
thismodeltorealdataforvarioustraitsofmultiplediseases,focusingondiseaseseverityanddrug
response.
Categories: QuantitativeTraitAnalysis
C6
MitochondrialGWAanalysisinseveralcomplexdiseasesusingtheKORA
population
AntoniaFlaquer1,Karl‐HeinzLadwig2,RebeccaEmeny2,MelanieWaldenberger3,HaraldGrallert3,
StephanWeidinger4,ChristaMeisinger5,ThomasMeitinger6,AnnettePeters2,KonstantinStrauch7
1InstituteofMedicalInformatics,BiometryandEpidemiology,ChairofGeneticEpidemiology,Ludwig‐
Maximilians‐Universität,Munich,Germany.
2InstituteofEpidemiologyII,HelmholtzZentrumMünchen‐GermanResearchCenterforEnvironmental
Health,Neuherberg,Germany
3ResearchUnitofMolecularEpidemiology,HelmholtzZentrumMünchen‐GermanResearchCenterfor
4DepartmentofDermatology,AllergologyandVenerology,UniversityHospitalSchleswig‐Holstein,Campus
Kiel,Kiel,Germany
5MyocardialInfarctionRegistry,Augsburg,Germany
6InstituteofHumanGenetics,HelmholtzZentrumMünchen‐GermanResearchCenterforEnvironmental
Maximilians‐Universitüt,Munich,Germany
MutationsofmitochondrialDNA(mtDNA)areunderagrowingscientificspotlight;scientistsbelieve
thesemutationsplayacentralroleinmany,ifnotmost,humandiseases.ThesmallcircularmtDNA
hasproventobeaPandora’sboxofpathogenicmutationsandrearrangements.Beingextremely
sensitivetoenvironmentalthreats,mitochondriaproducehigh‐energymolecules–adenosine
triphosphate(ATP).Mitochondriaalsogeneratereactiveoxygenspecies(ROS),whichparticipatein
cellsignalingandcommunication,particularlybetweennuclearandmitochondrialgenes.Ourmain
goalistoidentifymitochondrialsusceptibilitygenesforhumancomplexdiseases.Theclassical
statisticaltechniquesusedtodatetoanalyzethenucleargenomearenotappropriatetodirectlybe
appliedtothemitochondrialgenome.Someadjustmentsandnewmethodsneedtobedevelopedin
thecontextofmappingmitochondrialpolymorphisms.Usingdifferentgenotypingplatformssuchas
theAffymetrix6.0GeneChiparray,IlluminaMetaboChip200K,IlluminaHumanExomeBeadchip
array,andAffymetrixAxiomchiparrayweperformedmitochondrialGWAanalysusintheKORA
populationwithseveralphenotypes:BMI,cholesterol,post‐traumaticstressdisorder,thyroid
diseases,anxiety,depression,andasthma,amongothers.Ourfindingshighlighttheimportantrole
ofthemtDNAamongthefactorsthatcontributetotheriskofhumancomplexdiseasesandsuggest
thatvariantsinthemitochondrialgenomemaybemoreimportantthanhaspreviouslybeen
suspected.
Categories: Association:Genome‐wide,Association:UnrelatedCases‐Controls,Case‐ControlStudies,
Causation,PsychiatricDiseases,QuantitativeTraitAnalysis
C7
AdramaticresurgenceoftheGIGOsyndromeinthe21stcentury
FrançoiseClerget‐Darpoux1,EmmanuelleGénin2
1IHUimagine‐INSERMU781,Paris,France
2INSERMUMR1078,Brest,France
Inthesearchofthegeneticfactorsunderlyingmultifactorialdiseases,thewayispavedbyepidemics
ofGIGO(Garbage‐InGarbage‐Out)syndrome.Afirstoutbreaktookplaceinthelate1980’swhen
geneticistsbuildingonthesuccessofmodel‐basedlinkageanalysisinmonogenicdiseasesstartedto
usemonogenicmodelstostudymultifactorialdiseases.Asecondoutbreakisongoing,withgenome‐
wideassociationstudy(GWAS)heritabilityestimates.AlmostallGWASonmultifactorialdiseases
quantifythecontributionoftheidentifiedgeneticvariantstodiseasesusceptibilitythrough
heritabilityestimates.Theseestimatesarecomparedtotheonesobtainedfromfamilialdisease
segregationinordertodeterminehowmuchoftheheritabilityismissingandtopromptthesearch
forotherculpritssuchasrarevariants.Heritabilityestimatesareobtainedundertheadditive
polygenicmodelassumingthatthegeneticsusceptibilityinmultifactorialdiseaseisonlyexplained
byvariantswithmoderateandadditiveeffects.Thissimplisticmodelcannotberejectedbasedon
theinformationprovidedbybi‐allelictag‐SNPs,notbecauseitisthetruemodel,butbecausethis
informationisextremelypoorformodellingtheeffectofgeneticriskfactors.Severalexamplessuch
asPTPN22inrheumatoidarthritisclearlyillustratethefactthatusingthetag‐SNPinformation
alonemayleadtoahugeunderestimationoftherealeffectandtoanincorrectclassificationin
termsofrisk.GWAShasproventobeanefficienttoolforsusceptibilitygenedetectionbutnotfor
theirmodelling.Inthiswork,weshowhowheritabilityestimatescouldbebiasedwhenthedisease
modelismisspecified.
Categories: Association:Genome‐wide,Heritability,MultifactorialDiseases,PredictionModelling
C8
LargeScalePredictionandDissectionofComplexTraits
HaeKyungIm1,EricRGamazon1,KestonAquino‐Michaels1,NancyJCox1
1TheUniversityofChicago
Highaccuracypredictionofdiseasesusceptibilityanddrugresponseisnecessarytomake
personalizedorprecisionmedicineareality.Despiteinitialoptimismatthecompletionofthe
humangenomesequence,accuratepredictivetestsformanycommonconditionsarestill
unavailable.Thesmallportionofthetotalvariabilityexplainedbygenomewidesignificantgenes
havedampenedtheenthusiasm.However,studiesofthetotalheritabilityexplainedbyfullsetof
genotypedvariantsshowthatthereisampleroomforimprovement.Recentpowercalculations
haveshowedthatinordertoachievepredictionRsquaresclosetoheritabilityestimates,wemay
needmillionsofindividualsinourstudies.However,giventherateofincreaseinsamplesizesof
large‐scalemeta‐analysisstudies(overaquarterMillionforBMI),wearenottoomanyyearsaway
fromachievingthesenumbers.Also,advancesinelectronicmedicalrecordsandscalablecomputing
systemsareallowingustogatherandhandlethesemassivesamplesizes.Totakefulladvantageof
thegrowingamountofinformation,wearebuildingapubliclyavailablecatalogofprediction
models‐‐predictDB‐‐thathostsadditivemodelsforarangeofphenotypessuchasinflammation
markers,diseaserisk,lipidtraits,anthropomorphictraits,tonameafew.Furthermore,wehave
builtpredictionmodelsforgeneexpressionlevelsinmultipletissuesaswellasmicroRNAs.In
additiontoprediction,weusethesemodelstodissectthebiologyofcomplextraits.Forexample,we
usethepredictionmodelsforgeneexpressiontofindgenesthataredifferentiallyexpressedinsilico
betweencasesandcontrolsforarangeofdiseases.Thisisanovelgenebasedassociationtest,
termedPrediXcan,whichdirectlyteststhehypothesisthatgeneticvariationaltersdiseaserisk
throughtheregulationofgeneexpressionlevels.ApplicationtotheWellcomeTrustCaseControl
Consortiumdatayieldedmanygenome‐widesignificanthits.Manyofthemareknowndiseasegenes
butmanyarenovelandreplicationeffortsareunderway.
Categories: Association:Genome‐wide,GeneExpressionPatterns,PredictionModelling
C9
Geneticpredictorsoflongertelomeresarestronglyassociatedwithrisk
ofmelanoma
JenniferHBarrett1,DavidTBishop1,NicholasKHayward2,ChristopherIAmos3,PaulDPPharoah4,
FlorenceDemenais5,MatthewHLaw2,MarkMIles1,TheGenoMELConsortium
1UniversityofLeeds,UK
2QIMRBerghoferMedicalResearchInstitute,Brisbane,Australia
3DartmouthCollege,Hanover,USA
4UniversityofCambridge,UK
5INSERM,Paris,France
Telomeresprotectthesingle‐strandedchromosomeendsfromdamage.Telomeresshortenwithage
andenvironmentalexposuressuchassmoking.Telomerelength(TL)hasbeenrelatedtoanumber
ofage‐relateddiseases,usuallythroughcross‐sectionalstudiesfromwhichthedirectionofeffect
cannotbeinferred.Incontrasttomostdiseases,modestevidencehasaccumulatedthatlongerTLis
positivelyassociatedwiththenumberofmelanocyticneviandwiththeriskofafewcancers,
includingmelanoma.AsmoreisdiscoveredaboutthegeneticbasisofTL,Mendelianrandomisation
principlesmaybeinvokedtoelucidatethis.Arecentgenome‐wideassociationstudyofTLidentified
7genome‐widesignificantSNPs1.BasedontheseSNPsandtheirestimatedeffectsizesa“telomere
score”wascreated,anditsrelationshipwithmelanomariskwasinvestigatedusing>11,000cases
and13,000controls.Fourofthe7SNPsshowednominalevidenceofassociationwithmelanoma
risk(p<0.05).Therewasstrongevidenceofanassociationbetweenthescoreandmelanomarisk
(p<10‐8);theestimatedriskofmelanomatothosewithatelomerescoreinthehighestquartilewas
almost30%higherthantothosewithascoreinthelowestquartile.Furtheranalysissuggeststhat
whenthetelomerescoreusedhereisrefined,byusingadenserimputationpanelandbyincluding
moreSNPs,thescoreislikelytobeanevenstrongerpredictorofmelanomarisk.Thegenetic
associationsuggeststhat,ratherthanreversecausation,theassociationsobservedbetweenTLand
cancerriskaredueeithertoadirectcausaleffectoflongertelomeresortothepleiotropiceffectofa
numberofgenes.
1Coddetal,NatGenet2013;45:422‐427
Categories: Association:UnrelatedCases‐Controls,Cancer,MendelianRandomisation
C10
DetectionofcisandtranseQTLs/mQTLsinpurifiedprimaryimmune
cells
SilvaKasela1,2,LiinaTserel3,TõnuEsko4,Harm‐JanWestra5,LudeFranke5,KristaFischer4,Andres
Metspalu1,2,PärtPeterson3,LiliMilani4
1EstonianGenomeCenter,UniversityofTartu,Tartu,Estonia
2InstituteofMolecularandCellBiology,UniversityofTartu,Tartu,Estonia
3InstituteofGeneralandMolecularPathology,UniversityofTartu,Tartu,Estonia
5DepartmentofGenetics,UniversityMedicalCenterGroningen,UniversityofGroningen,Groningen,The
Netherlands
AdiverserepertoireofTcellsiscrucialforeffectivedefenseagainstinfectionwithpathogens
throughoutlife.CD4+Tcellsarevitalelementsoftheadaptiveimmuneresponse,whichhavebeen
associatedwiththepathogenesisofautoimmuneandinflammatorydiseases.CD8+Tcellsare
criticallyinvolvedindefenseagainstinfectionsandcanalsocontributetotheinitiationand
regulationofseveralorgan‐specificautoimmunediseases.Inordertoinvestigatethecell‐type
specificeffectsofnearbySNPsongeneexpression(ciseQTLs)andDNAmethylation(cismQTLs),
wepurifiedCD4+andCD8+cellsfromtheperipheralbloodofover600healthyindividuals.We
determinedtheSNPgenotypes(700K),expressionlevelsof47,000transcriptsfrom300subjects
andmethylationlevelsof450,000CpGsitesfromthe50youngestand50oldestsubjects.Intotal,
wedetectedmoreciseQTLsandmQTLsinCD4+comparedtoCD8+cellswithalargeoverlap
betweenthecellpopulations.Further,weselectedasetof9648SNPswhichhavebeenassociated
withimmunesystemrelateddiseasesfromstudiesusingtheImmunoChipandSNPsfromthe
reportsintheGWAScatalog.Despitetheseveralfoldsmallersamplesize,wewereabletoidentify
recentlyreportedtrans‐actingexpressionmasterregulatorSNPsonchromosome12and16(Fairfax
etal.2012,Westraetal.2013).Moreover,ourstudyrevealedthatsomeoftheeQTLsidentifiedin
wholebloodoriginatefromCD4+cellsonly,andwealsoidentifieddownstreamregulatedgenesthat
couldnotbedetectedinwholeblood.Forexample,wefoundthreeSNPsassociatedwithtype1
diabetes,Crohn’sdisease,andinflammatorybowel’sdiseasetoaffecttheexpressionoftheSTAT1
andIRF1genesintransinCD4+cells.
Categories: EpigeneticData,Epigenetics,GeneExpressionPatterns,GenomicVariation,Quantitative
TraitAnalysis
C11
WhyNext‐GenerationSequencingStudiesMayFail:Challengesand
SolutionsforGeneIdentificationinthePresenceofFamilialLocus
Heterogeneity
SuzanneMLeal1,RegieLynPSantos‐Cortez1,AtteeqURehman2,MeghanCDrummond2,Saima
Riazuddin3,DeborahANickerson4,WasimAhmad5,SheikhRiazuddin6,ThomasBFriedman2,EllenS
Wilch7
1CenterforStatisticalGenetics,DepartmentofMolecularandHumanGenetics,BaylorCollegeofMedicine,
Houston,Texas77030,USA
2LaboratoryofMolecularGenetics,NationalInstituteonDeafnessandOtherCommunicationDisorders,
NationalInstitutesofHealth,Rockville,Maryland20850,USA
3LaboratoryofMolecularGenetics,DivisionofPediatricOtolaryngologyHeadandNeckSurgery,Cincinnati
ChildrensHospitalMedicalCenter,Cincinnati45229,Ohio,USA
4UniversityofWashingtonCenterforMendelianGenomics
5DepartmentofBiochemistry,FacultyofBiologicalSciences,Quaid‐i‐AzamUniversity,Islamabad45320,
Pakistan
6NationalCenterofExcellenceinMolecularBiology,UniversityofthePunjab,Lahore54590,Pakistan
7DepartmentofMicrobiologyandMolecularGenetics,MichiganStateUniversity,EastLansing,Michigan
48824,USA
Next‐generationsequencing(NGS)ofexomesandgenomeshasacceleratedtheidentificationof
genesinvolvedinMendelianphenotypes.However,manyNGSstudiesfailtoidentifycausal
variants.Animportantreasonforsuchfailuresisfamiliallocusheterogeneity,wherecausalvariants
intwoormoregeneswithinasinglepedigreeunderlieMendeliantraitetiology.Asexamplesof
intra‐andinter‐sibshipfamiliallocusheterogeneity,wepresent10consanguineousPakistani
familiessegregatinghearingimpairment(HI)duetohomozygousmutationsintwodifferentHI
genesandalargeEuropean‐AmericanpedigreeinwhichHIiscausedbypathogenicvariantsin
threedifferentgenes.Wehaveidentified41additionalpedigreeswithsyndromicand
nonsyndromicHIforwhichasingleknownHIgenehasbeenidentifiedbutonlysegregateswiththe
phenotypeinasubsetofaffectedpedigreemembers.Weestimatethatlocusheterogeneityoccursin
15.3%(95%confidenceinterval11.9to19.9%)ofthefamiliesinourcollectionwherewehave
identifiedatleastonevariantinapreviouslypublishedHIgenewhichonlysegregateswithHI
phenotypeinasubsetofaffectedpedigreemembers.Wedemonstratenovelapproachestoapply
linkageanalysisandhomozygositymappingwhichcanbeusedtodetectlocusheterogeneityusing
eitherNGSorSNParraydata.Resultsfromtheanalysiscanalsobeusedtogroupsibshipsor
individualsmostlikelytobesegregatingthesamecausalvariantsandtherebyaidingene
identification.TheresultscanbeusedtoaidintheselectionofpedigreemembersforNGS.Itis
demonstratedhowthesemethodscanincreasethesuccessrateofgeneidentificationforfamilies
withlocusheterogeneity.
Categories: Association:Family‐based,Heterogeneity,Homogeneity,LinkageAnalysis,SequencingData
C12
Variationinestimatesofkinshipobservedbetweenwhole‐genomeand
exomesequencedata
ElizabethEBlue1
1UniversityofWashington
Genotypicvariationmaybeusedtoestimaterelationshipsbetweenindividuals.Theserelationships
areclearlyimportantwhenconfirmingpedigreestructureandtestingtheco‐segregationofa
variantwithatrait.Itisalsoimportantwhentestingassociationofageneticvariantwith
case/controlstatusinasetof“unrelated”subjects:ex.,thereasonwhyprincipalcomponentsare
includedascovariatestominimizetheeffectsofpopulationstratification.Thepopularityofexome
sequencingfordiseasegenediscoverysuggestsweneedtoknowwhetherthesedataprovide
accurateestimatesofrelationshipsbetweensubjects.Theexomerepresents~1%ofthegenome,
anddoesnotrepresentarandomsubsampleofgenomicvariation.Here,wecompareamethod‐of‐
momentsandtheKING‐robustestimatorofkinshipappliedtoSNPchipdata,wholeexome,and
wholegenomesequencedataforfoursubjectswithknownpedigreerelationships.SNPchip‐based
estimatesofkinshiparesimilartothepedigree‐basedexpectation,withtheKING‐robustestimates
deviatingslightlymorethanthemethod‐of‐momentsestimates.However,theexome‐based
estimatesaremuchmorevariable:overestimatingsomerelationshipsbyasmuchasathirdand
underestimatingothersbynearlyaquarterofthepedigree‐basedexpectation.Weexplorethe
effectsofallelefrequency,linkagedisequilibrium,andthenumberofmarkersonestimatesof
kinshipdrawnfromwholegenomesequencedata.Theseresultssuggestwemustaccountforthe
non‐randomdistributionofvariationintheexomewhenestimatingrelationshipsbetweensubjects.
Categories: Ascertainment,GenomicVariation,LinkageandAssociation,PopulationStratification,
SequencingData
C13
Robustgenotypecallingfromverylowdepthwholegenomesequencing
data
ArthurLGilly1,JeremySchwartzentruber1,AngelaMatchan1,Aliki‐EleniFarmaki2,George
Dedoussis2,PetrDanecek1,LorraineSoutham1,3,EleftheriaZeggini1
1WellcomeTrustGenomeCampus,Hinxton,CambridgeshireCB101SA,UK
2DepartmentofNutritionandDietetics,SchoolofHealthScienceandEducation,HarokopioUniversity,Athens,
Greece
3WellcomeTrustCentreforHumanGenetics,OxfordOX37BN,UK
Low‐depthwhole‐genomesequencing(WGS)hasbeenproposedasapowerfulapproachtocomplex
traitassociationstudydesign,asitallowsforareducedper‐samplecostandhencegreatersample
size.Howevervariantsandgenotypescalledatthesedepthstendtobelessreliablethanchip‐typed
ones,androbustguidelinesforvariantfiltering,genotyperefinementandimputationhavenotbeen
establishedyet.Inthisstudy,wefocuson995samplesfromaGreekisolatedpopulation(HELIC
study),sequencedatverylowdepth(1x).WeusedGWASandexomechipdata(availableforall
samples)andhigh‐depthexomesequencing(forasubsetofsamples)astruthsetsforperformance
calculations.WefindthattheVariantQualityScoreRecalibrationtooltypicallyusedtofilterlow‐
confidencesitescanreactunpredictablytosmallchangesintheunderlyingmodel’sparametersfor
lowdepthWGSdatacalling.Weshowthatthesepitfallscanbeavoidedwithacomprehensive
explorationoftheparameterspace.Wedemonstratethatover80%oftruelow‐frequency
(1%<MAF<5%)variantsarefound,comparedtoanaverage60%for0.1%<MAF<1%and40%for
MAF<0.1%.WeperformextensivebenchmarkingoftheBEAGLE,IMPUTE2andMVNCallrefinement
toolsandshowthatwiththehelpofthe1000Genomesreferencepanel,itispossibletoreacha
>95%genotypeconcordanceanda>90%minoralleleconcordanceacrossthewholeMAFspectrum.
Wereplicateknownassociationhits,therebyprovidingaproofofconceptforarobustprocessing
pipelineforlow‐depthWGSvariantcalls.
Categories: Association:Genome‐wide,Bioinformatics,DataMining,DataQuality,SequencingData
C14
Insightsintothegeneticarchitectureofanthropometrictraitsusing
wholegenomesequencedata
IoannaTachmazidou1,GrahamRSRitchie1,2,JosineMin3,KlaudiaWalter1,JieHuang1,JohnPerry4,
ThomasKeane1,ShaneMcCarthy1,YasinMemari1
1WellcomeTrustSangerInstitute,Hinxton,Cambridge,UK
2EuropeanMolecularBiologyLaboratory,EuropeanBioinformaticsInstitute,Cambridge,UnitedKingdom
3MRCIntegrativeEpidemiologyUnit,UniversityofBristol
4MRCEpidemiologyUnit,UniversityofCambridge
Bodyweightandfatdistributionmeasuresareassociatedwithincreasedriskofcardiometabolic
disease.AspartoftheUK10Kstudy,weinvestigatedthegeneticarchitectureof12anthropometric
traitsin3,538individualswith~7xwholegenomesequence(WGS)datafromtheALSPACand
TwinsUKcohorts.VariantsdiscoveredthroughWGS,alongwiththosefromthe1000Genomes
Project,wereimputedintoadditionalindividualsfromtheALSPACandTwinsUKcohortswith
GWASdata,increasingthetotalsamplesizeto11,178.Weinvestigatedassociationbetween
anthropometrictraitsand~9millionvariantswithMAF≥0.01and~5millionvariantswithMAF
0.001‐0.01.Insilicoreplicationwassoughtin16externalcohortsforatotalsamplesizeof15,000‐
40,000dependingontrait.Weobserveasignificantexcessofindependentpreviouslynotreported
variantswithMAF>0.01andp<10‐5inUK10Kinallanthropometrictraits.Wefindsignificant
enrichmentofvariantsassociatedwithBMIinUK10Kandestablishedmonogenicobesitygenes.
Furtherreplicationisongoing,butinterimanalysesidentifyreplicatingsignals,forexample,variant
chr5:105105444(EAF0.0084;UK10Kp=4.69x10‐5;replicationp=2.53x10‐4;overallp=5.53x10‐8,
samplesize=27,687)isanovelsignalassociatedwithwaistcircumferenceadjustedforBMI.Waist
tohipratioisassociatedwithvariantchr9:23016057(EAF0.003;UK10Kp=6.11x10‐5;replication
p=2.92x10‐04;overallp=5.98x10‐8,samplesize=25,373).Thesereplicatingsignalsareatvariants
withMAF<0.01,havemodesteffectsizesandarenotpresentinHapMap.Largersamplesizesare
requiredfortheidentificationandreplicationoffurtherrarevariantassociationswith
anthropometrictraits.
Categories: Association:Genome‐wide,QuantitativeTraitAnalysis,SequencingData
C15
StandardImputationversusGeneralizationsoftheBasicCoalescentto
EstimateGenotypes
MariaKabisch1,UteHamann1,JustoLorenzoBermejo2
1MolecularGeneticsofBreastCancer,GermanCancerResearchCenter(DKFZ),Heidelberg,Germany
2InstituteofMedicalBiometryandInformatics,UniversityofHeidelberg,Heidelberg,Germany
Genotypesthathavenotbeendirectlymeasuredareoftenimputedinassociationstudiestoincrease
statisticalpower,torefineassociationmapping,andtodetectgenotypingerrors.Mostoftenapplied
imputationmethodsexploitthepresentlinkagedisequilibrium(LD)amonggeneticvariantstoinfer
genotypes.ThiscausesastrongdependenceofimputationaccuracyonthesimilarityofLDpatterns
inthestudypopulationandthereferencepanel.Alternatively,coalescenttheoryassumesthat
haplotypesarerelatedthroughtheunderlyingpopulationgenealogy.Coalescent‐basedimputation
relaxestheassumptionofidenticalLDpatternsandmaythusresultinanincreasedaccuracy.To
examinethishypothesis,wefirstassessedtheimputationaccuracyunderthebasiccoalescent.
Studyandreferencehaplotypesweresimulatedusing'msms'[1].Haplotypeswerepairedatrandom
tomimicbiallelicvariants.Tenpercentofthevariantsinthestudywererandomlyselectedand
assumedtobedirectlymeasured,therestwasmasked.'BATWING'wasusedtoestimateone
thousandgenealogicaltrees,whichweresubsequentlysummarizedinaconsensusthreewith
'SumTree'[2,3].Expectedcoalescencetimeswereusedtoidentifyhaplotypetemplatesforgenotype
imputation.Finally,maskedgenotypeswereimputedandcomparedwiththetruegenotypesto
quantifytheaccuracyofimputation.Afterexamininggenotypeimputationunderthebasic
coalescent,populationgrowthandpopulationstructurewereincorporated.Imputationaccuracies
reachedbystandardmethods,e.g.IMPUTE2,willbecomparedwithcoalescent‐basedresultsatthe
IGES2014conference.
[1]Ewing,Hermisson(2010)MSMS:acoalescentsimulationprogramincludingrecombination,
demographicstructureandselectionatasinglelocus.Bioinformatics,26:2064‐5.
[2]Wilson,Weale,Balding(2003)InferencesfromDNAdata:populationhistories,evolutionary
processesandforensicmatchprobabilities.JournaloftheRoyalStatisticalSociety,SeriesA,166:
155‐88.
[3]Sukumaran,Holder(2010)DendroPy:APythonlibraryforphylogeneticcomputing.
Bioinformatics26:1569‐1571.
Categories: CoalescentTheory
C16
Improvementofgenotypeimputationaccuracythroughintegrationof
sequencedatafromasubsetofthestudypopulation
BarbaraPeil1,MariaKabisch2,ChristineFischer3,UteHamann2,JustoLorenzoBermejo1
2MolecularGeneticsofBreastCancer,GermanCancerResearchCenter(DFKZ),Heidelberg,Germany
3InstituteofHumanGenetics,UniversityHospitalHeidelberg,Heidelberg,Germany
Unmeasuredgenotypesingeneticassociationstudiescanbeestimated(imputed)usingexternal
datarepositories,forexampletheHapMap,ideallycomplementedwithsequencedatafromown
studyindividuals.Severalstudieshaveevaluatedwhichindividualsaremosthelpfulforgenotype
imputation.Initialeffortsfocusedonaselectionofreferenceindividualswhobestreflected
recombinationpatternsinthestudypopulation.Morerecently,theadvantageofgeneticdiversityin
thereferencepanelhasbeenrecognized.Wehavecompareddifferentstrategiestoselectstudy
individualsforsequencinginordertomaximizeimputationaccuracy.Fivealternativestrategies
wereexaminedinHapMapbasedsimulations.Thestrategy“none”incorporatednoadditional
sequencetotheexternalreferencepanel.Thestrategy“random”incorporatedthesequencesofa
randomsubsetof10%studyindividuals.Thestrategies“univariatedepth”,“bivariatedepth”and
“trivariatedepth”reliedonagenomewideprincipalcomponentanalysisofthestudypopulation,
followedbytheidentificationof10%ofstudyindividualswiththelargeststatisticaldepthbasedon
thefirstone,firsttwoandfirstthreeprincipalcomponents.Asexpected,theinclusionofadditional
sequencesfromtheownstudypopulationoutperformedimputationexclusivelyrelyingonexternal
referencepanels.Theselectionofstudyindividualsbasedontheunivariatedepthwasthebest
strategyinsimulationsmimickingEuropeanassociationstudies.Detailedresultsforadditional
investigatedscenarioswillbeprovidedattheconference.
Categories: Association:Genome‐wide,DataQuality,MissingData,SequencingData
C17
LearningGeneticArchitectureofComplexTraitsAcrossPopulations
MarcCoram1,SophieICandille1,HuaTang1
1StanfordUniversity
Genome‐wideassociationstudies(GWAS)havesuccessfullyrevealedmanylocithatinfluence
complextraitsanddiseasesusceptibilities.Anunansweredquestionis“towhatextentdoesthe
geneticarchitectureunderlyingatraitoverlapbetweenhumanpopulations?”Weexplorethis
questionusingbloodlipidconcentrationsasamodeltrait.Wedemonstratestrikingsimilaritiesin
geneticarchitectureoflipidtraitsacrosshumanpopulations.Inparticular,wefoundthata
disproportionatefractionoflipidvariationinAfricanAmericansandHispanicAmericanscanbe
attributedtogenomiclociexhibitingstatisticalevidenceofassociationinEuropeans,eventhough
theprecisegenesandvariantsremainunknown.Atthesametime,wefoundsubstantialallelic
heterogeneitywithinsharedloci,characterizedbothbypopulation‐specificrarevariantsand
variantssharedamongmultiplepopulationsthatoccuratdisparatefrequencies.Exploitingthis
overlappinggeneticarchitecture,wedevelopapopulation‐sensitiveapproachthatsubstantially
improvestheefficiencyofGWASinnon‐Europeanpopulations.
Categories: Association:Genome‐wide,Association:UnrelatedCases‐Controls,PopulationStratification
C18
Genome‐widegenotypeandsequence‐basedreconstructionofthe
140,000yearhistoryofmodernhumanancestry
DanielShriner1,FasilTekola‐Ayele1,AdebowaleAdeyemo1,CharlesNRotimi1
1NationalHumanGenomeResearchInstitute
Weinvestigatedancestryof3,528modernhumansfrom163ethno‐linguisticgroups.Weidentified
19ancestralcomponents,with94.4%ofindividualsshowingmixedancestry.Afterusingwhole
genomesequencestocorrectforascertainmentbiasesingenome‐widegenotypedata,wedatedthe
mostrecentcommonancestorto140,000yearsago.WedetectedanOut‐of‐Africamigration
100,000–87,000yearsago,leadingtopeoplesoftheAmericas,eastandnorthAsia,andOceania,
followedbyanothermigration61,000–44,000yearsago,leadingtopeoplesoftheCaucasus,Europe,
theMiddleEast,andsouthAsia.Wedatedeightdivergenceeventsto33,000–20,000yearsago,
coincidentwiththeLastGlacialMaximum.Werefinedunderstandingoftheancestryofseveral
ethno‐linguisticgroups,includingAfricanAmericans,Ethiopians,theKalash,LatinAmericans,
Mozabites,Pygmies,andUygurs,aswellastheCEUsample.Ubiquityofmixedancestryemphasizes
theimportanceofaccountingforancestryinhistory,forensics,andhealth.
Categories: Ascertainment,PopulationGenetics
C19
ModelComparisonandSelectionforCountDatawithExcessZerosin
MicrobiomeStudies
WeiXu1,2,AndrewDPaterson2,3,WilliamsTurpin4,KennethCroitoru4,LizhenXu3
1DepartmentofBiostatistics,PrincessMargaretHospital,Toronto,ON,Canada
2PrograminGeneticsandGenomeBiology,theHospitalforSickChildren,Toronto,ON,Canada
3DallaLanaSchoolofPublicHealth,UniversityofToronto,Toronto,ON,Canada
4DivisionofGastroenterology,ZaneCohenCentreforDigestiveDiseases,MountSinaiHospital,Toronto,ON,
M5T3L9,Canada
Inhumanmicrobiomestudies,itisoftenofinteresttoidentifyclinicalorgeneticfactorsthatare
associatedwithdifferentbacterialtaxa.Themicrobiotasequencecountdataarecomplexwith
featuressuchashighdimension,over‐dispersion,andoftenexcesszeros.Inaddition,thenumberof
totalreadsvariesamongsubjects.Zeroinflatedorhurdlemodelsprovidepossibleanalytic
approachesforthistypeofdataandthevariationintotalreadscanbeadjustedasoffsets.However,
inpractice,onepartmodelswhichignorezeroinflationareoftenused.Todeterminethepatternof
superiorityofusingzeroinflatedorhurdlemodelsoverthesimplifiedonepartmodels,wedesigned
extensivesimulationstudiestocomparetheperformanceofdifferentstatisticalmethodsundera
varietyofgeneratingscenarios.Thesescenariosinclude:differentlevelsofzeroinflation;presence
ofdispersion;differentmagnitudeanddirectionsofthecovariateeffectonboththestructuralzero
andcountcomponents.Theresultsshowthat,comparedtoone‐partmodels,thehurdleandzero
inflatedmodelshavewellcontrolledtypeIerrors,higherpower,bettergoodnessoffitmeasures,
andaremoreaccurateandefficientintheparameterestimation.Besidesthat,thehurdlemodels
havesimilargoodnessoffitandparameterestimationforthecountcomponentastheir
correspondingzeroinflatedmodels.However,theestimationandinterpretationfortheparameters
forthezerocomponentscanbedifferent.Inaddition,wedevelopedacomprehensivemodel
selectionandanalysisstrategytoanalyzethistypeofdata.Thisstrategywasimplementedinagut
microbiomestudyof>400independentsubjects.
Categories: MicrobiomeData,QuantitativeTraitAnalysis
C20
BayesianLatentVariableModelsforHierarchicalClusteredTaxaCounts
inMicrobiomeFamilyStudieswithRepeatedMeasures
LizhenXu1,AndrewDPaterson1,2,WeiXu2,3
1DallaLanaSchoolofPublicHealth,UniversityofToronto,Toronto,Canada
2PrograminGeneticsandGenomeBiology,theHospitalforSickChildren,Toronto,ON,Canada
3DepartmentofBiostatistics,PrincessMargaretHospital,Toronto,ON,Canada
Inmicrobiomestudies,taxacountdataareoftenover‐dispersedandincludeexcesszeros.
Furthermore,differenttaxabelongingtothesametaxonomichierarchicalclusterareoften
correlatedduetotheirsimilar16SrRNAsequences.Addedcharacteristicsofmicrobiomedataisthe
repeatedmeasuresonrelatedfamilymembers.Jointmodelingofmultipletaxausingfamilydata
withrepeatedmeasuresisdesirablebutnon‐trivialduetothecomplexcorrelationandmulti‐
dimensionaloutcomedata.Toovercomethesechallenges,weproposetousethelatentvariable
(LV)methodology.TheLVapproachlinksthemultipletaxacountsbyintroducingalatentrandom
variablethatrepresentstheunobservedtraitoftheircommontaxonomycluster.Thelatentvariable
formulationalsoprovidesaflexiblewaytoallowforoutcomeswithdiscretecomponents,inour
case,thenegativebinomialoutcomeswithorwithoutzeroinflation.LValsoprovidesaneffective
waytodetectpleiotropicgenes,witheffectsonmultipletaxa.WebuildourLVinferenceina
Bayesianframework.SamplingsfromtheposteriordistributionareobtainedusingMCMC
algorithms.Theparameterexpansiontechniqueisusedtoimprovethemixingofchainsandthe
Bayesiandevianceinformationcriteria(DIC)andBayesfactorsareusedformodelselection.
Extensivesimulationsshowthatourmethodperformswellincapturingthecorrelationsamongthe
multipletaxainducedbysharedhostgeneticfactors.Wethenillustrateourmethodwithagut
microbiomestudyofleanandobesetwins.
Categories: BayesianAnalysis,MarkovChainMonteCarloMethods,MicrobiomeData,Multivariate
Phenotypes,QuantitativeTraitAnalysis
C21
ARetrospectiveLikelihoodApproachforEfficientIntegrationofMultiple
OmicsandNon‐OmicsFactorsinCase–ControlAssociationStudiesof
ComplexDiseases
BrunildaBalliu1,RoulaTsonaka1,StefanBoehringer1,JeanineHouwing‐Duistermaat1
1LeidenUniversityMedicalCenter,TheNetherlands
Integrativeomics,thejointanalysisofoutcomeandmultipletypesofomicsdata,suchasgen‐omic,
epigen‐omicandtranscript‐omicdata,offersapromisingalternativetogenome‐wideassociation
studies,formorepowerfulandbiologicallyrelevantassociationstudies[1,2].Thesestudiesusually
employthecase‐controldesign,andtheyoftenincludedataonadditionalnon‐omiccovariates,e.g.
ageorgender,thatmaymodifytheunderlyingomicsriskofcasesorcontrols.Anunanswered
questionishowtobestintegratemultipleomics,andpossiblynon‐omicsinformationtomaximize
statisticalpowerinstudiesthatascertainindividualsonthebasisofphenotype.Mostpublications
onintegrativeomicshavereliedonsomevariantoftheprospectivelogisticregressiontomodelthe
associationbetweenoutcomeandriskfactors[2].However,whilesuchanapproachhasimproved
powerinstudieswithrandomascertainment,relativetomethodsthatanalyzeeachdatasource
separately,itoftenlosespowerundercase‐controlascertainment[3].Inthisarticle,weproposea
novelstatisticalmethodforintegratingmultipleomics,andpossiblynon‐omicsfactors,incase‐
controlassociationstudies.Ourmethodisbasedonaretrospectivelikelihoodfunctionthatproperly
reflectsthecase‐controlsampling,bymodelingthejointdistributionoftheomicsandnon‐omics
factorsconditionalonthecase‐controlstatus.Whenpossible,weexplicitlyimposethe
independenceassumptionbetweentheomicsandnon‐omicscovariates.Thenewmethodprovides
accuratecontroloffalse‐positiverateswhilemaximizingstatisticalpower.Themethodisillustrated
usingsimulatedandrealdataexamples.
[1]HLi(2013),WIRE:SBM,5(6):677–686.
[2]HuangYetal.(2014),Ann.Appl.Stat.,8(1):352‐376.
[3]Zaitlenetal.(2012),PLoSGenet8(11):e1003032.
Categories: Ascertainment,Case‐ControlStudies,DataIntegration,EpigeneticData,Epigenetics,Gene
ExpressionArrays,MaximumLikelihoodMethods
C22
Inferenceforhigh‐dimensionalfeatureselectioningeneticstudies
ClausTEkstrøm1
1Biostatistics,UniversityofCopenhagen
Featureselectionisanecessarystepinmanygeneticapplicationsbecausethebiotechnological
platformsprovideacheapandfastmeansforproducinghigh‐dimensionaldata.Thisneedfor
dimensionreductionisheightenedfurtherforexamplewhendatafromdifferentomicsare
combinedintosimultaneousintegrateddataanalysisorwhenhigher‐levelinteractionsamongthe
availablepredictorsareconsidered(whichisthecaseforgene‐geneorgene‐environment
interactionsorinepigenetics).PenalizedregressionmodelssuchastheLassoortheelasticnethave
provedusefulforvariableselectioninmanygeneticapplications‐especiallyforsituationswith
high‐dimensionaldatawherethenumbersofpredictorsfarexceedsthenumberofobservations.
Thesemethodsidentifyandrankvariablesofimportancebutdonotgenerallyprovideanyinference
oftheselectedvariables.Thus,thevariablesselectedmightbethemost''important''butneednot
besignificant.Weproposeasignificancetestforevaluatingthenumberofsignificantselection(s)
foundbytheLasso.Thismethodrephrazesthenullhypothesisandusesarandomizationapproach
whichensuresthattheerrorrateiscontrolledevenforsmallsamples.Theabilityofthealgorithm
tocomputep‐valuesoftheexpectedmagnitudeisdemonstratedwithsimulateddataandthe
algorithmisappliedtotwodataset:oneonprostatecancerandafullGWAS.Theproposedmethod
isfoundtoprovideapowerfulwaytoevaluatethesetofselectionsfoundbypenalizedregression
whenthenumberofpredictorsareseveralordersofmagnitudelargerthanthenumberof
observations.
Categories: Association:Family‐based,Association:Genome‐wide,Bioinformatics,Gene‐Environment
Interaction,Gene‐GeneInteraction
Posters
P1
Increasedpowerfordetectionofparent‐of‐origin(imprinting)effectsin
genome‐wideassociationstudiesusinghaplotypeestimation
RichardHowey1,HeatherJCordell1
1NewcastleUniversity,UK
Ingeneticstudies,parent‐of‐origin(imprinting)effectscanbeconsideredasthephenomenon
wherebyanindividual’sphenotypedependsbothontheirowngenotypeandontheparentalorigin
oftheconstituentalleles.Severalmethodshavebeenproposedtodetectsucheffectsinthecontext
ofstudiesofcase/parenttrioswithsinglenucleotidepolymorphism(SNP)genotypedata.Formost
case/parenttrios,thegenotypecombinationsaresuchthattheparent‐of‐originoftheallelesinthe
childcanbedeterminedunambiguously,butthisisnottruewhenallthreeindividualsare
heterogenousatasingleSNPunderstudy.Existingmethodsforthedetectionofparent‐of‐origin
effectsinthecontextofgenome‐wideassociationstudies(GWAS)thuseitherperformsomesortof
“averaging”overpossibleconfigurationsorelsediscardtheseambiguouscase/parenttrios.The
powertodetectparent‐of‐origineffectswouldbeincreasedifthetrueparentaloriginofthealleles
couldbedeterminedwithahigherdegreeofcertainty.WepresenthereanextensiontotheGWAS
methodimplementedinthePREMIM/EMIMsoftwaretodetectparent‐of‐origineffectsusing
externalestimatesofhaplotypesprovidedbytheprogramSHAPEIT2,therebyusingsurrounding
SNPinformationtohelpbetterestimatetheparentaloriginofallelesatagiventestSNP.Weshow
throughsimulationsthatourapproachhasincreasedpoweroverpreviousversionsofEMIMand
achievespowerneartothatachievediftheparent‐of‐originofalleleswereknown.
P2
EpidemiologicalProfileofCleftPalateintheStateofBahia‐Brazil
MarcelaMQLLeiro1,RenataLLFdeLima1,LuziaPolianadosAnjosSilva1
1UniversityFederalofBahia,Brazil
Cleftlipandpalate(FLPs)areasetofmalformationsofthefacerepresentingthemostcommon
congenitalanomaliesofthehumanspecies.Braziliandataoncraniofacialanomaliesarestill
consideredscarceandscattered,duetothedifficultyofreportingthesecasesinthepublichealth
system.Giventhedifferentpopulation,environmental,social,lifestyleandissuesofracial
miscegenationinBrazilcharacteristics,theprevalenceofthisanomalyseemstovaryineachstate
ofthecountry.OBJECTIVE:TodescribetheepidemiologyofpatientswithCleftLipand/orPalate
residentsoftheStateofBahia.Studyofquantitativetraitrunsthroughcross‐sectionalcaseseries
withsamplegroupconsistedofchildrenaged0‐12years,whoarepartofaprogramofcarein
Centrinho‐BA.RESULTS:Ofthe206patientstherewasaslightprevalenceoffemales(51%),
andnon‐syndromiccases,95%.Ofthetotalsample,53%hadCLPandonly19%FL,and119
cases(58%)wereborninthestate.TheFLPwasmoreprevalentinpatientswithapositivefamily
history,71cases(34.5%).RegardingtheetiologyofPLF9.3%(19cases)reportedhavingused
alcoholduringpregnancy.Itwasnotedsocioeconomicsituationofvulnerabilityinpatientswith
CLPwhere60%(124cases)hadanincomeof1‐3minimumwages.CONCLUSION:Itwas
observedthroughthisstudy,ahigherincidenceofCLPinrelationtoFLassociatedwithahigher
prevalenceinblacks,withthesocioeconomicvulnerabilityexposedpopulation.
P3
GeneralizedFunctionalLinearModelsforGene‐basedCase‐Control
AssociationStudies
RuzongFan1,YifanWang1,JamesLMills1,ToniaCCarter2,IrynaLobach3,AlexanderFWilson4,Joan
EBailey‐Wilson4,DanielEWeeks5,MomiaoXiong6
1NationalInstituteofChildHealthandHumanDevelopment,NationalInstitutesofHealth
2MarshfieldClinic
3UniversityofCalifornia,SanFrancisco
4NationalHumanGenomeResearchInstitute,NationalInstitutesofHealth
5UniversityofPittsburgh
6UniversityofTexas‐Houston
Byusingfunctionaldataanalysistechniques,wedevelopedgeneralizedfunctionallinearmodelsfor
testingassociationbetweenadichotomoustraitandmultiplegeneticvariantsinageneticregion
whileadjustingforcovariates.Bothfixedandmixedeffectmodelsareproposedandcompared.
ExtensivesimulationsshowthatRao'sefficientscoretestsoftheproposedfixedeffectmodelsare
veryconservativesincetheygeneratelowtypeIerrors,andglobaltestsofthemixedeffectmodels
areveryrobustsincetheygenerateaccuratetypeIerrors.Furthermore,wefoundthattheRao's
efficientscoreteststatisticsoftheproposedfixedeffectmodelshavehigherpowerthanthe
sequencekernelassociationtest(SKAT)anditsoptimalunifiedversion(SKAT‐O)inmostcases
whenthecausalvariantsarebothrareandcommon.Whenthecausalvariantsareallrare(i.e.,
minorallelefrequencieslessthan0.03),theRao'sefficientscoreteststatisticsandtheglobalscore
testshavesimilarorslightlylowerpowerthanSKATandSKAT‐O.Inpractice,itisnotknown
whetherrarevariantsorcommonvariantsinagenearedisease‐related.Allwecanassumeisthata
combinationofrareandcommonvariantsinfluencesdiseasesusceptibility.Thus,thesuperior
performanceoftheproposedmodelswhenthecausalvariantsarebothrareandcommonshows
thattheproposedmodelscanbeveryusefulindissectingcomplextraits.SNPdatarelatedtoneural
tubedefectsandHirschsprung'sdiseaseareanalyzedbytheproposedmethodsandSKATand
SKAT‐Oforarealapplicationandcomparison.Themethodscanbeusedineithergene‐disease
genome‐wide/exome‐wideassociationstudiesorcandidategeneanalyses.
Categories: Association:CandidateGenes,Association:Genome‐wide,Association:UnrelatedCases‐
Controls,Case‐ControlStudies,LinkageandAssociation,MultilocusAnalysis,MultipleMarker
DisequilibriumAnalysis,SequencingData
P4
Geneticanalysisofthechromosome15q25.1regionidentifiesIREB2
variantsassociatedwithlungcancer
ChristopherIAmos1,IvanPGorlov1,JamesDMcKay2,LoïcLeMarchand3,YafangLi1,Gianluca
Severi4,DavidCChristiani5,PaulBrennan2,JohnKField6,RayjeanJHung7
1DartmouthCollege
2InternationalAgencyforResearchonCancer
3UniversityofHawaii
4HumanGeneticsFoundation,Torino,ItalyandUniversityofMelbourne
5HarvardUniversitySchoolofPublicHealth
6UniversityofLiverpool
7UniversityofToronto
Genome‐wideassociationstudiesoflungcanceridentifiedtheregionofchromosome15q25.1that
includesanicotinicacetylcholinereceptorclusterasbeingthemoststronglyassociatedwithlung
cancerrisk.Tocharacterizetheimpactthatspecificfunctionalvariantsinthisregionhaveuponrisk
forlungcancerdevelopmentweperformedfinemappingselectingallcurrentlyknownSNPs
influencinglungcancerriskalongwithcodingSNPsinthe200megabaseregionsurrounding
CHRNA5,ageneknowntoinfluencesmokingbehaviorinthisregion.Markersusedinanalysis
wereselectedbaseduponthefollowingcriteria:knownfunctionaleffectonactivity,validationin
AfricanorEuropeanpopulations,positionacrosstheregion,predictedeffectonfunction,r‐square
withothermarkerslessthan80%.Wefinemappedtheregionbygenotyping1395SNPsextending
fromthegeneCRABP1toADAMTS7fromposition79103132toposition79103132usingacustom
AffymetrixAxiomarrayin3063casesand2940controlsofEuropeanancestryfrom5studies:MSH‐
PMH,EPIC,MEC,LLPC,HPFS&NHS.Oddsratios(OR)adjustedforage,sex,thefirsttwoprincipal
componentsandpopulationwereestimatedusinglogisticregression.Acrossthisregion,101SNPs
metthemultipletestingcorrectedthreshold(p<3.5×10‐5).ThemostsignificantSNPslieinaregion
ofIREB2withthemostsignificantlyassociatedvariantbeingrs17483686(OR=1.26,p=8.93x10‐
12).ThepreviouslywellcharacterizedSNPinCHRNA5,rs16969968,whichcausesreduced
signaling,yieldedalesssignificantassociation(OR=1.24,p=8x10‐10).Thesefindingssuggest
IREB2,agenerelatedtoironmetabolism,playsaroleinlungcancerdevelopmentinadditionto
nearbynicotinicreceptors.
Categories: Association:CandidateGenes,Cancer,FineMapping,Gene‐EnvironmentInteraction
P5
Anovelintegratedframeworkforlargescaleomicsassociationanalysis
RamounaFouladi1,2,KyryloBessonov1,2,FrancoisVanLishout1,2,JasonHMoore3,KristelVanSteen1,2
1SystemsandModelingunit,MontefioreInstitute,UniversityofLiege,Liege,Belgium
2BioinformaticsandModeling,GIGA‐R,UniversityofLiege,Liege,Belgium
3DepartmentofGenetics,InstituteforQuantitativeBiomedicalsciences,GeiselSchoolofMedicineat
Dartmouthcollege,Lebanon,US
Genome‐wideassociationstudies(GWAstudies)havebeenverysuccessfulinidentifyingnumerous
geneticlociassociatedwithawiderangeofcomplextraits.Thesediscoverieshaverevealednew
pathwaysthatseemtoplayasignificantroleincommondiseases.Singleomicsstudies,suchas
GWAs,onlyprovidelimitedinformationtodisease‐relatedbiologicalorfunctionalmechanisms.In
anomics–diseasetraitassociationsetting,ideally,agenerictooliscreatedthatcandealwith
differentgranularitiesofomicsinformation(i.e.,differentarchitecturesofcommonandrare
variants,epigeneticmarkers,geneexpression).Here,anovelomicsassociationanalysistechniqueis
proposedthatbuildsupontheModel‐BasedMultifactorDimensionalityReduction(MB‐MDR)
framework.Atthebasisofthemethodliesadataorganizationstepthatinvolvesclusteringof
individuals.InthefirstimplementationsofMB‐MDR,thesefeatureswereSNPs,andindividuals
wereclusteredaccordingtotheirgenotypes.IngenomicMB‐MDR,anyfeature(continuousor
categorical)canbeanalyzed,andfeaturesmappedtogenomic“regionsofinterest”(ROIs)are
submittedtoaclusteringalgorithmtofindgroupsofsimilarindividualsonthebasisofselected
ROIs.Whenappliedtoexome‐sequencingdata,wecanidentifyageneasaROI,andcantakeboth
rareandcommonfeaturesmappedtotheseregionsasinputfeatures.Wethenproposetocluster
individualsaccordingtotheirsimilaritiesbasedonrareandcommonvariants,afterwhichclassic
MB‐MDRisapplied.Theperformanceofseveralfeatureselectionmethods,similaritymeasures,and
clusteringalgorithmsingenomicMB‐MDRisinvestigatedusingsyntheticandreal‐lifeexome
sequencingdata.
Categories: Association:CandidateGenes,Bioinformatics,DataIntegration,Epigenetics
P6
InclusiveCompositeIntervalMappingandSkew‐NormalDistribution
ElisabeteFernandes1
1CEMAT‐CenterforComputacionalandStochasticMathematics,Portugal
Thecompositeintervalmapping,CIM,(JansenandStam,1994;Zeng,1994)isthemostcommonly
usedmethodforQTLmappingwithpopulationsderivedfrombiparentalcrosses.However,theCIM
maynotcompletelyensureallitsadvantageousproperties.Themodifiedalgorithm,calledas
inclusivecompositeintervalmapping,ICIM,(Wangetal.,2007)hasasimplerformthanthatusedin
CIM,butafasterconvergencespeed.ICIMretainsalladvantagesofCIMoverIMandavoidsthe
possibleincreaseofsamplingvarianceandthecomplicatedbackgroundmarkerselectionprocessin
CIM.Thisapproachmakesuseoftheassumptionthatthequantitativephenotypefollowsanormal
distribution(KruglyakandLander,1995).Manyphenotypesofinterest,however,followahighly
skeweddistribution,andinthesecasesthefalsedetectionofamajorlocuseffectmayoccur
(Morton,1984).Aninterestingalternativeistoconsideraskew‐normalmixturemodelinICIM,and
theresultingmethodisheredenotedasskew‐normalICIM.Thismethod,whichissimilartoICIM,
assumesthatthequantitativephenotypefollowsaskew‐normaldistributionforeachQTLgenotype.
Themaximumlikelihoodestimatesofparametersoftheskew‐normaldistributionareobtainedby
theexpectation‐maximization(EM)algorithm.Theproposedmodelisillustratedwithrealdata
fromanintercrossexperimentthatshowsasignificantdeparturefromthenormalityassumption.
Theperformanceoftheskew‐normalICIMisassessedviastochasticsimulation.Theresultsindicate
thattheskew‐normalICIMhashigherpowerforQTLdetectionandbetterprecisionofQTLlocation
ascomparedtoICIM.
Categories: Association:CandidateGenes,Association:Genome‐wide,FineMapping,Maximum
LikelihoodMethods,QuantitativeTraitAnalysis
P7
Transmission‐basedTestsForGeneticAssociationUsingSibshipData
HemantKulkarni1,SaurabhGhosh1
1IndianStatisticalInstitute,Kolkata
TheclassicalTransmissionDisequilibriumTest(TDT)forbinarytraits(Spielmanetal.1993)isa
family‐basedalternativetopopulationbasedcase‐controlstudiesandisprotectedagainst
populationstratification,andhence,anassociationfindingcanbeattributedtothepresenceof
linkage.TherehavealsobeensomeextensionsoftheclassicalTDTforquantitativetraits.However,
thesetests,whicharebasedonthetriodesign(twoparentsandanoffspring)donotremainvalidas
testsforassociationinthepresenceofsibshipdatasincethemarginaleffectoflinkagecanresultin
transmissionbiasofalleles.Inourstudy,wehavemodifiedtheTDTtestprocedureforbothbinary
aswellasquantitativetraitsbasedonsibshipdatausingapermutationbasedapproach.Weselect
oneoffspringatrandomfromeachfamilyandcomputetheusualtrio‐basedteststatistic.Werepeat
thisprocedureandconsidertwoteststatisticsbasedonthemeanandthemaximumvalueofthe
trio‐basedteststatisticsobtainedoverdifferentreplications.Weobtaintheexactdistributionofthe
teststatisticusingpermutations.Weperformextensivesimulationstoevaluatethepowersofthe
proposedtestsunderawidespectrumofgeneticmodelsanddifferentdistributionsofa
quantitativetrait.Wefindthattheteststatisticbasedonthemeanyieldsmorepowercomparedto
thatbasedonthemaximum.
Categories: Association:CandidateGenes
P8
Identificationofrarecausalvariantsinsequence‐basedstudies
MarinelaCapanu1,IulianaIonita‐Laza2
1MemorialSloanKetteringCancerCenter
2ColumbiaUniversity
Pinpointingthesmallnumberofcausalvariantsamongtheabundantnaturallyoccurringgenetic
variationisadifficultchallenge,butacrucialoneforunderstandingprecisemolecularmechanisms
ofdiseaseandfollow‐upfunctionalstudies.Weproposeandinvestigatetwocomplementary
statisticalapproachesforidentificationofrarecausalvariantsinsequencingstudies:abackward
eliminationprocedurebasedongroupwiseassociationtests,andahierarchicalapproachthatcan
integratesequencingdatawithdiversefunctionalandevolutionaryannotationsforindividual
variants.Usingsimulations,weshowthatincorporationofmultiplebioinformaticpredictorsof
deleteriousness,suchasPolyPhen‐2,SIFTandGERP++scores,canimprovethepowertodiscover
trulycausalvariants.Asproofofprinciple,weapplytheproposedmethodstoVPS13B,agene
mutatedintherareneurodevelopmentaldisordercalledCohensyndrome,andrecentlyreported
withrecessivevariantsinautism.Weidentifyasmallsetofpromisingcandidatesforcausal
variants,includingarare,homozygousprobably‐damagingvariantthatcouldcontributetoautism
risk.
Categories: Association:CandidateGenes,Case‐ControlStudies,GenomicVariation,SequencingData
P9
TargetedresequencingofGWASloci:insightintogeneticetiologyofcleft
lipandpalatethroughanalysisofrarevariantswithfocusonthe8q24
region
MargaretATaub1,ElizabethJLeslie2,TheCleftSeqConsortium
1JohnsHopkinsUniversity
2UniversityofPittsburgh
Non‐syndromiccleftlipwithorwithoutcleftpalate(CL/P)isacommonbirthdefectwithcomplex
inheritance.Despiteconsiderableprogressinidentifyingrisklociinseveralgenome‐wide
associationstudies(GWAS),identificationofthecausalvariantsateachlocusremainsachallenge.
Tothisend,weselectedthirteenregionsfromearlierGWASandcandidategenestudies,totaling
6.3Mb,fortargetedcaptureanddeepsequencingin1521case‐parenttrioswithCL/Pfromseveral
populations.Weperformedstatisticalanalysesoncommon,denovoandrarevariants.Here,we
focusonthelatter,inparticularinthe8q24region.Whilemanyrarevarianttestsfocusoncoding
variants,8q24,asagenedesert,requiresotherapproaches.Weperformedregulatory‐regionbased
burdenteststoseeifrarevariantsinaparticularregulatoryelementwereover‐orunder‐
transmitted.Noresultsweresignificantaftermultipletestingcorrection.Weusedthelikelihood‐
ratiobasedScan‐Triomethodtofindwindowswithover‐orunder‐transmittedrarevariants,
restrictingouranalysestovariantswithCADDscore>10andassessingsignificancebypermuting
transmittedanduntransmittedhaplotypes.Thisanalysisrevealedapromisingclusterofvariants
neartheGWAShitin8q24.Wealsodidhaplotype‐basedtestingwherehaplotypesweregroupedby
allelecarriedatrs72728755,theSNPgivingmostsignificantsignalinthetransmission‐
disequilibriumtest(TDT).Wetestedfordifferencesinthepresenceofrarevariantsbetween
deleteriousandprotectivehaplotypesbysearchingslidingwindowsforclustersofrarevariants
seenonlyontransmittedhaplotypes.Significancewasevaluatedbypermutation.Grants:U01‐
HG005925;R01‐DE016148.
Categories: Association:CandidateGenes,Association:Family‐based,HaplotypeAnalysis,Linkageand
Association,SequencingData
P10
Ajointassociationmodelofeffectsofrareversuscommonvariantson
Age‐relatedMacularDegeneration(AMD)usingaBayesianhierarchical
generalizedlinearmodel
WilmarMIgl1,fortheInternationalAMDGenomicsConsortium(IAMDGC)
1DepartmentofGeneticEpidemiology,UniversityofRegensburg,Germany
Purpose:AMDisacommoncauseofblindnessinolderpeoplewithastronggeneticcontribution
fromcommonvariants(CVs).Recently,severalrarevariants(RVs,MAF<1%)werefound.Sofar
thecontributionofRVsandCVshasnotbeenexaminedinacomprehensivejointmodel.Methods.
TheIAMDGCdatacomprise33,976unrelatedEuropeans(16,144AdvancedAMDcases,17,832
controls).569,645variantsacrossthegenomeweregenotypedonacustom‐modified
HumanCoreExomearraybyIllumina.Theanalysesfocuson18knownand17novellocifromsingle‐
variantanalyses.TheappliedBayesianhierarchicalgeneralizedlinearmodel(here:logistic,Yiand
Zhi,2011)extendsthegeneralizedlinearmodelframeworkbyjointlyestimatingindividualvariant
andgroupvariant(here:rarevs.common)effectsbasedongeneticriskscores.Weaklyinformative
Bayesianpriors(HierarchicalCauchy)wereused.Allresultswereadjustedforancestryprincipal
componentsandDNAsourceascovariatesandformultipletestingperlocus.Results.Theanalysis
of225rareversus199commonvariants(total424,α=1E‐4)intheCFIlocus,showedindependent
group‐leveleffectsofrare(OR=2.06,CI95%=[1.92;2.22],p=1.09E‐87)andcommon(OR=2.39,
CI95%=[2.18,2.62],p=7.38E‐77)variants.Significantsinglevarianteffectswereonlyobservedfor
theknownrarevariantrs141853578(G119R,OR=3.24,CI95%=[2.06;5.10],p=3.34E‐07)inthis
jointmodel.Resultsforotherlociwillbepresented.Conclusions.Jointmodelingofgeneticeffects
giveadditionalinsightsintothegeneticarchitectureofdiseasecomparedtoconventionalsingle‐
varianttests.ReferencesYi,N.,&Zhi,D.(2011).Bayesiananalysisofrarevariantsingenetic
associationstudies.GeneticEpidemiology,35(1),57–69.
Categories: Association:CandidateGenes,Association:UnrelatedCases‐Controls,BayesianAnalysis,
Case‐ControlStudies
P11
AssociationBetweenBloodPressureSusceptibilityLociandUrinary
Electrolytes
BamideleOTayo1,HollyKramer1,ColinAMcKenzie2,GuichanCao1,RamonDurazo‐Arvizu1,Amy
Luke1,TerrenceForrester2,RichardSCooper1
1LoyolaUniversityChicago,Maywood,IL
2UniversityoftheWestIndies,Kingston,Jamaica
BACKGROUND:Genome‐wideassociationstudieshaveledtoidentificationandvalidationofabout
40susceptibilitylociforbloodpressureandhypertensionespeciallyamongindividualsofEuropean
ancestry.Eventhoughthesegeneticvariantscollectivelyexplainonlyasmallfractionofthe
heritabilityforbloodpressurephenotypes,similarassociationswithbloodpressurephenotypes
remaintobedemonstratedinindividualsofAfricanancestry.OBJECTIVE:Aspartofthestudyon
geneticsofhypertensioninBlacks,wesoughttoidentifypossibleassociationsbetweenBP
susceptibilitylociandurinarysodiumandpotassiumamongindividualsofAfricanorigin.METHOD:
Weobtainedmeandailyurinarysodiumandpotassiumfromthree24‐hoursamplescollectedfrom
613adultJamaicansthatconsistedof140malesand473females.Thesubjectsweregenotyped
usingtheIlluminaMetaboChipgenotypingarraythatcontainsselectedvariantsformetabolicand
atherosclerotic/cardiovasculardiseasetraits.Inthepresentstudy,weanalyzedonlytheavailable
qualitycontrolled25bloodpressuresusceptibilitylocipreviouslyreportedbyTheInternational
ConsortiumforBloodPressure.Eachofthe25variantswastestedforassociationwithurinary
sodiumandpotassiumunderanadditivegeneticmodeofinheritanceusingmultivariablelinear
regressionmodelthatadjustedforage,sex,bodymassindexandage‐by‐sexinteractioncovariates.
Tocontrolforpossiblepopulationstratificationfromadmixture,wealsoincludedthefirst10
principalcomponentsfromtheautosomalgenotypesinthemodel.RESULTS&CONCLUSION:Our
findingsrevealassociation(p<0.004)betweenurinarypotassiumandvariantsintheTBX5‐TBX3
(rs10850411),ADM(rs7129220)andPLCE1(rs932764)loci.Thisstudyprovidespreliminarydata
thatgeneticvariantsassociatedwithBPsusceptibilitymaybeassociatedwithurinarysodiumand
potassiumexcretion;additionalstudiestoconfirmthesefindingsarerequired.
Categories: Association:CandidateGenes,CardiovascularDiseaseandHypertension
P12
Asystematicevaluationofshorttandemrepeatsinlipidcandidategenes:
ridingontheSNP‐wave
ClaudiaLamina1,MargotHaun1,StefanCoassin1,AnitaKloss‐Brandstätter1,ChristianGieger2,
AnnettePeters3,KonstantinStrauch2,LyudmylaKedenko4,BernhardPaulweber4,Florian
Kronenberg1
1DivisionofGeneticEpidemiology,DepartmentofMedicalGenetics,MolecularandClinicalPharmacology,
InnsbruckMedicalUniversity,Innsbruck,Austria
2InstituteofGeneticEpidemiology,HelmholtzZentrumMünchen‐GermanResearchCenterforEnvironmental
Health(GmbH),Neuherberg,Germany
3InstituteofEpidemiologyII,HelmholtzZentrumMünchen‐GermanResearchCenterforEnvironmental
4FirstDepartmentofInternalMedicine,ParacelsusPrivateMedicalUniversitySalzburg,Austria
Structuralgeneticvariantsasshorttandemrepeats(STRs)arenottargetedinSNP‐based
associationstudiesandthus,theirpossibleassociationsignalsaremissed.Wesystematically
searchedforSTRsingeneregionsknowntocontributetototalcholesterol,HDLcholesterol,LDL
cholesterolandtriglyceridelevelsintwoindependentstudies(KORAF4,n=2553andSAPHIR,
n=1648),resultingin16STRsthatwerefinallyevaluated.Inacombineddatasetofbothstudies,the
sumofSTRalleleswasregressedoneachphenotype,adjustedforageandsex.Theassociation
analyseswererepeatedfor1000GimputedSNPsina200kbregionsurroundingtherespectiveSTRs
intheKORAF4Study.ThreeSTRsweresignificantlyassociatedwithtotalcholesterol(withinLDLR,
theAPOA1/C3/A4/A5/BUD13generegionandABCG5/8),fivewithHDLcholesterol(3within
CETP,oneinLPLandoneinAPOA1/C3/A4/A5/BUD13),threewithLDLcholesterol(LDLR,
ABCG5/8andCETP)andtwowithtriglycerides(APOA1/C3/A4/A5/BUD13andLPL).Noneofthe
investigatedSTRs,however,showedasignificantassociationafteradjustingfortheleadoradjacent
SNPswithinthatgeneregion.TheevaluatedSTRswerefoundtobewelltaggedbytheleadSNP
withintherespectivegeneregions.Therefore,theSTRsreflecttheassociationsignalsbasedon
surroundingSNPs.Inconclusion,noneoftheSTRscontributedadditionallytotheSNP‐based
associationsignalsidentifiedinGWASonlipidtraits.
Categories: Association:CandidateGenes,QuantitativeTraitAnalysis
P13
Linkagedisequilibriummappingofmultiplefunctionallociincase‐
controlstudies
Yen‐FengChiu1,Li‐ChuChien1,Kung‐YeeLiang2,Lee‐MingChuang3
1NationalHealthResearchInstitutes,Taiwan,ROC
2NationalYangMingUniversity,Taiwan,ROC
3NationalTaiwanUniversityHospital,Taiwan,ROC
Mostcomplexdiseasesaremultifactorial,involvingmultiplegeneticfactorsandtheirjointeffects.
Forsuchdiseases,methodsaccountingformultiplelocimaybemorepowerfulthansingle‐locus
analysesandmayofferimprovedprecisionofdisease‐locuslocalization.Weproposea
semiparametricmultipointlinkagedisequilibrium(LD)mappingapproachtoestimate
simultaneouslythediseaseloci,thegeneticeffectsofdiseaseloci,andthejointeffectsand
interactionsoftwoadjacentloci,andtoconstructcorrespondingCIsfortheseparameters.This
proposedmethodbuildsuponlargesampleproperties,whichisusefulforahigh‐densitygenome‐
wideassociationstudy(GWAS)withcommonvariants.ChromosomalregionscanbedividedbyLD
blocksorgenestolocalizefunctionallociineachsubregion.Weapplytheproposedapproachtoa
dataexampleofcase‐controlstudies.Resultsofthesimulationsanddataexamplesuggestthatthe
developedmethodperformswellintermsofbias,variance,andcoverageprobabilityunder
scenarioswithuptothreediseaseloci.
Categories: Association:CandidateGenes,Association:Genome‐wide,Case‐ControlStudies,Gene‐Gene
Interaction,PopulationGenetics
P14
Geneticvariantsintransporterandmetabolizinggenesandsurvivalin
colorectalcancerpatientstreatedwithoxaliplatincombination
chemotherapy
ElisabethJKap1,PetraSeibold1,YesildaBalavarca2,LinaJansen1,NataliaBecker1,Michael
Hoffmeister1,CorneliaMUlrich2,BarbaraBurwinkel3,HermannBrenner1,JennyChang‐Claude1
1GermanCancerResearchCenter
2NationalCenterforTumorDiseases
3UniversityofHeidelberg
Oxaliplatinhasbecomeoneofthemainchemotherapeuticagentsforthetreatmentofcolorectal
cancer(CRC).Metabolicandtransporterenzymesareinvolvedintheclearanceofchemotherapeutic
agents.Variantsingenesencodingtheseenzymesmaycausevariationinresponsetooxaliplatin
andcouldthereforebepotentialpredictivemarkers.Thereforewecomprehensivelyassessed
differentialeffectsof931geneticvariantsintransporterandmetabolizinggenesandoverall
survival(OS)inCRCpatientswhoreceivedoxaliplatinchemotherapycomparedtopatientstreated
withotherchemotherapeutics.Weincluded623CRCpatientsdiagnosedbetween01.01.2003and
31.12.2007andrecruitedinaGermanpopulation‐basedstudy(DACHS),whoreceivedadjuvant
chemotherapy(201patientsreceivedoxaliplatin).SurvivalanalysiswasperformedusingaCox
regressionmodel,adjustedforage,sex,UICCstage,cancersiteandBMI.Effectmodificationby
oxaliplatintreatmentwasassessedusingamultiplicativeinteractionterm.Medianfollow‐uptimein
patientsreceivingoxaliplatinwas4.9yearsafterwhich96patientsweredeceased.Rs11203943
(NAT1),rs7017402(NAT1)andrs4148872(TAP2)showeddifferentialassociationwithOS
accordingtooxaliplatintreatment(Unadjustedp‐values<0.001),althoughresultswerenot
significantafterFDRcorrection(FDRp<0.05).OurdatasuggestthatgeneticvariantsinNAT1and
TAP2maybepredictivemarkersforoxaliplatintreatment.WeplantouseadditionalSNPs(imputed
tothe1000genomereferencepanel)toidentifyfurtherpotentialpredictivemarkers.
Categories: Association:CandidateGenes,Cancer
P15
Post‐Genome‐WideAssociationStudyUsingGeneralizedStructured
ComponentAnalysis
HelaRomdhani1,AurélieLabbe1,HeungsunHwang1
1McGillUniversity
Weareinterestedindevelopingastatisticalframeworkforthejointanalysisofmultiplecorrelated
traitsandmultiplegenotypemeasuresfromcandidateregionsingeneticstudies.Weproposetouse
structuralequationmodelingwithlatentvariablesfortheassociationstructurebetweenthe
observedvariablesandsomecomponentsmediatingtherelationshipsbetweengenotypesand
phenotypes.Themodelisconstructedonthebasisofpriorbiologicalknowledgeofbothclinicaland
geneticpathways.WeusetheGeneralizedStructuredComponentAnalysis(GSCA)toestimatethe
model'sparameters.TestproceduresfordifferentkindsofdirectedeffectsmeasuredbyGSCAhave
beendevelopedandpowershavebeenassessedbysimulations.Finally,ananalysisoftheQCAHS
surveydataisperformedusingthisnewapproach.
Categories: Association:CandidateGenes,MultivariatePhenotypes,Pathways
P16
DetectingMaternal‐FetalGenotypeInteractionsAssociatedwith
ConotruncalHeartDefects:AHaplotype‐basedAnalysiswithPenalized
LogisticRegression
MarioACleves1,MingLi1,SteveWErickson1,CharlotteAHobbs1,JingyunLi1,XinyuTang1,ToddG
Nick1,StewartLMacleod1
1UniversityofArkansasforMedicalSciences
Non‐syndromiccongenitalheartdefects(CHDs)developduringembryogenesisasaresultofa
complexinterplaybetweenenvironmentalexposures,geneticsandepigeneticcauses.Genetic
factorsassociatedwithCHDsmaybeattributedtoeitherindependenteffectsofmaternalorfetal
genes,ortheinter‐generationalinteractionsbetweenmaternalandfetalgenes.Detectinggene‐by‐
geneinteractionsunderlyingcomplexdiseasesisamajorchallengeingeneticresearch.Detecting
maternal‐fetalgenotype(MFG)interactionsanddifferentiatingthemfromthematernal/fetalmain
effectshaspresentedadditionalstatisticalchallengesduetocorrelationsbetweenmaternaland
fetalgenomes.Traditionally,geneticvariantsaretestedseparatelyformaternal/fetalmaineffects
andMFGinteractionsonasingle‐locusbasis.Weconductedahaplotype‐basedanalysiswitha
penalizedlogisticregressionframeworktodissectthegeneticeffectassociatedwiththe
developmentofnon‐syndromicconotruncalheartdefects(CTD).Ourmethodallowssimultaneous
modelselectionandeffectestimation,providingaunifiedframeworktodifferentiatematernal/fetal
maineffectfromtheMFGinteractioneffect.Inaddition,themethodisabletotestmultiplehighly
linkedSNPssimultaneouslywithaconfigurationofhaplotypes,whichreducesthedata
dimensionalityandtheburdenofmultipletesting.ByanalyzingadatasetfromtheNationalBirth
DefectsPreventionStudy(NBDPS),weidentifiedsevengenes(GSTA1,SOD2,MTRR,AHCYL2,GCLC,
GSTM3andRFC1)associatedwiththedevelopmentofCTDs.OurfindingssuggestthatMFG
interactionsbetweenhaplotypesin3of7genes,GCLC,GSTM3andRFC1,areassociatedwithnon‐
syndromicconotruncalheartdefects.
Categories: Association:CandidateGenes,Association:Family‐based,Association:UnrelatedCases‐
Controls,Gene‐GeneInteraction
P17
Mutationsscreeningofexons7and13ofTMC1gene(DFNB7/11)in
Iranianautosomalrecessivenon‐syndromichearingloss(NSHL)
probandsusingmoleculartechniques
PayamGhasemi‐Dehkordi1,NegarMoradipour1,FatemehHeibati2,ShahrbanuoParchami‐Barjui1,
AhmadRashki3,MortezaHashemzadeh‐Chaleshtori1
1CellularandMolecularResearchCenter,ShahrekordUniversityofMedicalSciences,Shahrekord,Iran
2ClinicalBiochemistryResearchCenter,ShahrekordUniversityofMedicalSciences,Sharekord,Iran
3FacultyofVeterinaryMedicine,DepartmentofPhysiopathology,ZabolUniversity,Zabol,Iran
Non‐syndromichearingloss(NSHL)isthemostcommonbirthdefectwhichoccurinapproximately
1/1000newborns.NSHLisaveryheterogeneoustraitandcouldbecausedduetobothgeneticand
environmentalfactors.Mutationsoftransmembranechannel‐like1(TMC1)genecausenon‐
syndromicdeafnessinhumansandmice.Theaimofpresentstudywastoinvestigatethe
associationofTMC1genemutationsoflocusDFNB7/11inexons7and13inacohortof100
patientswithhearinglossinIranusingpolymerasechainreaction‐singlestrandedconformation
polymorphism(PCR‐SSCP),heteroduplexanalysis(HA),andDNAsequencing.Thebloodsamplesof
hearinglosspatientswerecollectedfrom10provincesofIran.DNAwasextractedfromspecimens
andmutationsofexons7and13ofTMC1genewereinvestigatedusingPCR‐SSCP.Inaddition,all
sampleswerecheckedbyheteroduplexanalysis(HA)reactionandsuspectedspecimenswithshift
bandsweresubjectedtoDNAsequencingforinvestigatethepresenceofanygenevariation.Inthis
study,nomutationwasfoundinthesetwoexonsofTMC1gene.TheseresultsconcludedthatTMC1
genemutationshaveaverylowcontributioninpatientsandwerenotgreatclinicalimportancein
theseprovincesofIran.However,morestudiesareneedtoinvestigatetherelationshipbetween
otherpartsofthisgenewithhearinglossindifferentpopulationthroughthecountry.Moreresearch
couldclarifytheroleofthisgeneanditsrelationwithdeafnessandprovideessentialinformation
forthepreventionandmanagementofauditorydisordercausedbythisgeneinIranianpopulation.
Keywords:TMC1gene,Hearingloss,PCR‐SSCP,Heteroduplexanalysis,Iran
Categories: Association:CandidateGenes,GenomicVariation
P18
ConotruncalHeartDefectsandCommonVariantsinMaternalandFetal
GenesinFolate,HomocysteineandTranssulfurationPathways
MarioACleves1,CharlotteAHobbs1,StewartLMacLeod1,StephenWErickson1,XinyuTang1,Ming
LI1,JingyunLi1,NickTodd1,SadiaMalik1
1UniversityofArkansasforMedicalSciences
Congenitalheartdefects(CHDs)arethemostprevalentstructuralbirthdefect,occurringin8to11
ofevery1,000livebirths.Conotruncalheartdefects(CTDs)compriseasubgroupofCHDsthatare
malformationsofcardiacoutflowtractsandgreatarteries.Weinvestigatedtheassociationbetween
CTDsandmaternalandfetalsinglenucleotidepolymorphisms(SNPs)in60genesinthefolate,
homocysteineandtransulfurationpathways.Wealsoexaminedwhetherpericonceptionalmaternal
folicacidsupplementationmodifiedtheseassociations.ParticipantswereenrolledintheNational
BirthDefectsPreventionStudybetween1997and2007.DNAsamplesfrom616case‐parental
triadsaffectedbyCTDsand1,645control‐parentaltriadsweregenotypedusingacustomIllumina®
GoldenGateSNParray.Log‐linearhybridmodels,optimizingdatafromcaseandcontroltriads,
wereusedtoidentifymaternalandfetalSNPsassociatedwithCTDs.Wakefield'sBayesianfalse‐
discoveryprobabilitymethod(BFDP)wasusedtoidentifyingnoteworthyassociations.Among921
SNPs,17maternaland17fetalSNPshadaBFDP<0.8.Tenofthe17maternalSNPsand2ofthe17
fetalSNPswerefoundwithintheglutamate‐cysteineligase,catalyticsubunit(GCLC)gene.Fetal
SNPswiththelowestBFDPwerefoundwithinthethymidylatesynthetase(TYMS)gene.
Additionally,thegeneticriskofCTDsfor19maternaland9fetalSNPswasfoundtobemodifiedby
periconceptionalfolicaciduse.Theseresultssupportpreviousstudiessuggestingthatmaternaland
fetalSNPswithinfolate,homocysteineandtranssulfurationpathwaysareassociatedwithCTDrisk.
MaternaluseofsupplementscontainingfolicacidmaymodifytheimpactofSNPsonthedeveloping
heart.
Categories: Association:CandidateGenes,Association:Family‐based,Association:UnrelatedCases‐
Controls
P19
GeneticPredispositionofXRCC1inSchizophreniaPatientsofSouth
IndianPopulation
SujithaSP1,LakshmananS2,HarshavaradhanS3,GunasekaranS1,AnilkumarG1
1SchoolofBiosciencesandTechnology,VITUniversity,Vellore632014TamilNadu,India
2GovernmentVelloreMedicalCollege,Vellore,TamilNadu,India
3SriNarayaniHospitalandResearchCentre,Vellore,TamilNadu,India
Schizophreniaisadebilitatingneuropsychiatricdisorder.Severalofthepreviousstudiescarriedout
toexploretheetiologyofthischronicdiseasesuggestforitsassociationwiththeSNPs(including
thenon‐synonymousones)atvariousgeneloci;andtheseinvestigationsproducedvaryingresults
dependingonethnicity.RoleofXRCC1asarepairgenehasbeenextensivelystudiedonawide
varietyofcarcinomas.Wehavecompellingreasonstoconsiderthisasacandidategenethatcould
influenceschizophrenia.However,barringafewinstances,theassociationstudiesonschizophrenia
andtheSNPatXRCC1arequitemeager.Thepresentstudy,performedonatotalof523subjects
including260casesand263controls,depictstheassociationofrs25487(Arg399Gln)
polymorphismofXRCC1withschizophrenia.Theanalysisrevealedthestronggenotypic
(‘AA’/Gln399Gln;p=0.006)andallelic(‘A’/Gln399;p=0.003,OR=1.448;95%CI=1.132to1.851)
associationoftheSNPwithschizophrenia.Wearefurtherencouragedtoanalyzetheassociationof
nicotine(ifany)withschizophrenia,inasmuchastheindividualswithschizophreniahaveshown
highersusceptibilitytonicotineaddiction.Thisstudywasperformedintwocohortswith260case
subjects(101nicotinesubstanceaddictsand159nicotinesubstancenaïvesubjects)and263
controlsubjects(with90subjectswithaddictionand173subjectswithoutaddiction).Thestudydid
notshowanyassociationofnicotineaddictionwiththisnon‐synonymousmutation.Toconclude,
thepresentstudyclearlydemonstratedtheassociationofGln399GlnwithschizophreniainTamil
population,andhasruledouttheroleofnicotineinthepolymorphismasanepigeneticfactor
influencingthedisease.
Categories: Association:CandidateGenes,Epigenetics,PsychiatricDiseases
P20
Astochasticsearchthroughsmokingimagesinmovies,geneticand
psycho‐socialfactorsassociatedwithsmokinginitiationinMexican
Americanyouths
MichaelDSwartz1,MatthewDKoslovsky1,ElizabethAVandewater2,AnnaVWilkinson3
1UniversityofTexasSchoolofPublicHealth,DivisionofBiostatistics
2UniversityofTexasSchoolofPublicHealth,DivisionofHealthPromotionandBehavioralScience
3UniversityofTexasSchoolofPublicHealth,DivisionofEpidemiology,HumanGeneticsandEnvironmental
Science
Sincesmokingisoneofthestrongestriskfactorsforlungcancer,identifyingfactorsrelatedto
smokinginitiationcanhaveahighimpactonreducinglungcancerrates.Ethnicdifferencesin
initiationrateshavebeenobserved,andMexicanAmericanyouthshavebeenunderstudied.Recent
independentstudieshaveidentifiedmultiplefactorsassociatedwithsmokinginitiationinMexican
Americanyouths:exposuretosmokingimagesinmovies,genetic,andpsycho‐socialfactors.Here
wesimultaneouslyinvestigateallthesefactorsandtheirpotentialinteractions.Usingaprospective
cohortof1,328MexicanAmeicanyouths,weinvestigatedsinglenucleotidepolymorphisms(SNPs)
fromtheopioidreceptoranddopaminepathways,psycho‐socialfactorsandexposuretosmoking
relatedimagesinmovies.Wemeasuredpsycho‐socialfactorsusingpreviouslyvalidated
questionnairesandexposuretosmokingimagesinmoviesusingtheBeachmethod.Weused
stochasticsearchvariableselectionmethodologytojointlyassesstheseassociationswithsmoking
initiationinMexicanAmericanyouths.Weusedpriorsthatbothimposedhierarchicalmodelsfor
interactionsandcontrolledthefalsepositiverate.Ourpreliminaryfindingsidentifiedsmoking
imagesinmovies,age,gender,positiveoutcomeexpectationsfromsmoking,risktakingtendencies,
livingwithasmoker,peerinfluence,andservingdetentioninschool,andaSNPongeneSNAP25
andanotheronOPRM1relatedtosmokinginitiation.Wedidnotidentifyanyinteractions.
Categories: Association:CandidateGenes,BayesianAnalysis,Gene‐EnvironmentInteraction,Markov
ChainMonteCarloMethods
P21
AssociationbetweenApolipoproteinEgenotypeandcancer
susceptibility:ameta‐analysis
AnandR1,PrakashSS1,VeeramanikandanR1,RichardKirubakaran2
1ChristianMedicalCollege,Vellore,India
2SouthAsianCochraneCenter,ChristianMedicalCollege,Vellore,Tamilnadu,India‐632002.
ApolipoproteinE(ApoE),aproteinprimarilyinvolvedinlipoproteinmetabolismoccursin3
isoforms(E2,E3andE4).WhilestudiesevaluatingtheassociationbetweenApoEgenotypeand
incidenceofmalignanciesareavailable,theresultsareinconsistent.Theobjectiveofthepresent
studywastoanalyzetheassociationbetweenAPOEgenotypeandincidenceofcancerbyameta‐
analysis.Weconductedaliteraturesearchintheelectronicdatabasesforstudieswithinformation
onAPOEpolymorphismsinmalignancies.Sixteenstudies(14case‐control/2cohort;77970controls
and12010cases)wereincludedforthepresentmeta‐analysis.Pooledoddsratios(OR)with95%
confidenceintervals(CI)werecalculatedassumingarandom‐effectmodelforallthegenotypesand
alleles.Subgroupanalysesbasedonstudydesign,ethnicityofpopulations,andsiteofcancerand
sourceofcontrolswereperformedasapost‐hocmeasure.Appropriateteststodetect
heterogeneity,publicationbiasandsensitivityweredoneatallstages.Thepooledeffectmeasurefor
thecomparisonsdidnotrevealanassociationinprimaryanalyses.Inthesubgroupanalyseswe
observedasignificantnegativeassociationbetweenAPOE4+genotypesandoverallriskofcancerin
thecohortstudysubgroup.TherewasalsoaweakpositiveassociationbetweenAPOE4+genotypes
andbreastcancer.Weobservedamoderateinter‐studyheterogeneityforseveralofthe
comparisons(I2<40%).Sensitivityanalysesdidnotaltertheoverallpooledeffectmeasureinthe
majorcomparisons.Therewerenoevidencestosuggestapublicationbias.Overall,thepresent
meta‐analysisdidnotshowanyassociationbetweenAPOEallelesorgenotypeswithincidenceof
canceringeneral.
Categories: Association:CandidateGenes,Cancer,Case‐ControlStudies
P22
NovelapproachidentifiesSNPsinSLC2A10andKCNK9withevidencefor
parent‐oforigineffectonbodymassindex
CliveJHoggart1,GiuliaVenturini2,MassimoMangino3,FeliciaGomez4,GeorgeDavey‐Smith5,
ValentinRousson6,JoelNHirschhorn7,CarloRivolta1,RuthJFLoos8,ZoltanKutalik6
1DepartmentofGenomicsofCommonDisease,ImperialCollegeLondon,LondonW12ONN,UK
2DepartmentofMedicalGenetics,UniversityofLausanne,Lausanne1005,Switzerland
3DepartmentofTwinResearch&GeneticEpidemiology,King'sCollegeLondon,LondonSE17EH,UK
4DepartmentofGenetics,DivisionofStatisticalGenomics,WashingtonUniversitySchoolofMedicineinSt.
Louis,St.Louis63108,USA
5MRCIntegrativeEpidemiologyUnit,UniversityofBristol,BristolBS82BN,UK
6InstituteofSocialandPreventiveMedicine(IUMSP),CentreHospitalierUniversitaireVaudois(CHUV),
Lausanne1010,Switzerland
7CenterforBasicandTranslationalObesityResearchandDivisionsofEndocrinologyandGenetics,Boston
Childrenâ€™sHospital,Boston2115,USA
8MRC‐EpidemiologyUnit,UniversityofCambridge,CambridgeCB20QQ,UK
Thephenotypiceffectofsomesinglenucleotidepolymorphisms(SNPs)dependsontheirparental
origin.Wepresentanovelapproachtodetectparent‐of‐origineffects(POE)ingenome‐wide
genotypedataofunrelatedindividuals.Themethodexploitsincreasedphenotypicvarianceinthe
heterozygousgenotypegrouprelativetothehomozygousgroups.Weappliedthemethodto
>56,000unrelatedindividualstosearchforPOEsinfluencingbodymassindex(BMI).SixleadSNPs
werecarriedforwardforreplicationinfivefamily‐basedstudies(of~4,000trios).TwoSNPs
replicated:thepaternalrs2471083‐Callele(locatedneartheimprintedKCNK9gene)andthe
paternalrs3091869‐Tallele(locatedneartheSLC2A10gene)increasedBMIequally(beta=0.11
(SD),P<0.0027)comparedtotherespectivematernalalleles.Real‐timePCRexperimentsof
lymphoblastoidcelllinesfromtheCEPHfamiliesshowedthatexpressionofbothgeneswas
dependentonparentaloriginoftheSNPsalleles(P<0.01).Ourschemeopensnewopportunitiesto
exploitGWASdataofunrelatedindividualstoidentifyPOEsanddemonstratesthattheyplayan
importantroleinadultobesity.
Categories: Association:Family‐based,Association:Genome‐wide,Association:UnrelatedCases‐
Controls,Epigenetics,QuantitativeTraitAnalysis,TransmissionandImprinting
P23
InteractiveeffectbetweenDNAH9geneandearly‐lifetobaccosmoke
exposureinbronchialhyper‐responsiveness
Marie‐HélèneDizier1,RachelNadif2,PatriciaMargaritte‐Jeannin1,SheilaJBarton3,ValérieGagné‐
Ouellet4,ChloéSarnowski1,MyriamBrossard1,NolwennLavielle1,JocelyneJust5,MarkLathrop6
1INSERM,U946,UniversitéParisDiderot,Paris,France
2INSERM,U1018,Villejuif,UniversitéParisSud,France
3FacultyofMedicine,UniversityofSouthampton,Southampton,UK
4UniversitéduQuébec,Chicoutimi,Canada
5Centredel’AsthmeetdesAllergies,INSERM,UMR_S1136,EquipeEPAR,France
6McGillUniversity,Montréal,Canada
Wepreviouslyperformedagenome‐widelinkageanalysisofbronchialhyper‐responsiveness(BHR)
testinginteractionwithearlylifeenvironmentaltobaccosmoke(ETS)exposureintheFrench
EpidemiologicalstudyontheGeneticsandEnvironmentofAsthma(EGEA)Ourgoalwastoconduct
fine‐scalemappingofthedetected17p11regionthatshowedlinkageinETSunexposedsiblings
only,toidentifygeneticvariantsinteractingwithETSexposurethatinfluenceBHR.Analyseswere
firstperformedinthe388FrenchEGEAasthmaticfamilies,usingfamily‐basedassociationtest
(FBAT).TosearchforSNPxETSinteraction,weusedatwo‐stepstrategy:1)selectionofSNPs
showingFBATassociationsignalswithBHR(P<0.01)inunexposedsiblings;2)FBAThomogeneity
testbetweenexposedandunexposedsiblingsofselectedSNPs.ForSNPsshowingsignificant
interaction,alog‐linearmodelingapproachfortestinginteraction,asproposedbyUmbachand
Weinberg(2000),wasappliedforvalidation.Replicationanalyseswerethenconductedintwo
independentasthmaticfamilysamples:253French‐Canadianfamilies(SLSJ)and341UKfamilies.In
EGEAfamilies,17SNPsshowedassociationsignalswithBHRinunexposedsiblings.AsingleSNP
showedsignificantinteractionwithETSexposureusingbothmethods(P≤10‐3).Thisresultwas
replicatedintheSLSJfamiliesandmeta‐analysisofthetwosamplesprovidedastrongimprovement
inthedetectionofinteraction(P=7.10‐5).TherewashowevernoreplicationintheUKfamilies.
TheSNPshowingsignificantinteractiveeffectwithETSexposureinBHRisinapromisingcandidate
gene,DNAH9,agenewellknowntobeassociatedwithPrimaryCiliaryDyskinesia.Funded:ANR‐
GWIS‐AM‐2011,RégionIdF
Categories: Association:Family‐based,Gene‐EnvironmentInteraction,MultifactorialDiseases
P24
DetectionofrarehighlypenetrantrecessivevariantsusingGWASdata
StevenGazal1,MouradSahbatou2,Marie‐ClaudeBabron1,Jean‐CharlesLambert3,PhilippeAmouyel3,
EmmanuelleGénin4,Anne‐LouiseLeutenegger1
1INSERMU946,Paris,France
2CEPH,Paris,France
3INSERMU744,Lille,France
4INSERMU1078,Brest,France
Genome‐wideassociationstudies(GWAS)haveidentifiedseveralcommongeneticvariantsin
multifactorialdiseases.However,takentogether,thesevariantsonlyexplainasmallpartofthe
heritability.Differentcandidateshavebeensuggestedtoexplainthismissingheritability,and
amongthemarevariantswithrecessiveeffectsthatcouldplayarolebuthavenotbeendetectedso
far.Recessivevariantsareeasytodetectwhentheyarerare,fullypenetrant,andinvolvedinrare
monogenicdiseases.Thestrategyofchoicetodetectthemishomozygositymapping(HM),a
powerfulapproachthatconsistsinfocusingoninbredfamiliesandsearchingforaregionofthe
genomeofsharedhomozygosityintheinbredcases.Withthehelpofgenome‐widegeneticdata,itis
nowpossibletodetermineifanindividualisinbredbasedontheobservedgenomehomozygosity
patterns.HMcanthenbeperformedwithoutanyknowledgeofthegenealogy.Thiscouldbeused
notonlytodetectrarerecessivevariantsinvolvedinmonogenicdiseases,butalsotoidentify
recessiveMendeliansubentitiesofmultifactorialdisease.Severalsoftwarehavebeendevelopedto
studyinbreeding.However,noneofthemprovideanintegrativesolutiontoestimateinbreeding,
identifyandvisualizerunsofhomozygositybydescentandperformHM.Wehaverecently
developedtheFSuitepipelinetoopenupthepossibilitytoeasilydetectinbredcasesinGWAS
dataset,andtofocusonthemtoperformHMallowingforheterogeneity.Wewillillustratethe
possibilitiesofferedbyFSuiteonaFrenchGWASdatasetincluding1,886affectedindividualswith
Alzheimer’sdisease.About5%ofthecaseswerefoundinbredandwereeligibleforHM,allowing
thedetectionof3candidategenomicregions.
Categories: Association:Family‐based,Association:Genome‐wide,Inbreeding,IsolatePopulations,
MultifactorialDiseases
P25
CopyNumberVariation(CNV)detectioninwholeexomesequencingdata
forMendeliandisorders
PengZhang1,HuaLing1,ElizabethPugh1,KurtHetrick1,DaneWitmer1,NaraSobreira2,DavidValle2,
KimDoheny1
1CenterforInheritedDiseaseResearch,InstituteofGeneticMedicine,TheJohnsHopkinsSchoolofMedicine
2InstituteofGeneticMedicine,TheJohnsHopkinsSchoolofMedicine
TheCentersforMendelianGenomics(CMG)projectusesnext‐generationsequencingand
computationalapproachestodiscoverthegenesandvariantsthatunderlieMendelianconditions.
WhileSNVsandINDELsexplainsomeMendelianconditions,manyremainunresolved.Weare
interestedtoknowtowhatextentunrecognizedCNVswouldresolvesomeofthese.Comparedto
wholegenomesequencing(WGS),whole‐exomesequencing(WES)isacost‐effectivealternativefor
findingdiseasegenesharboringvariantswithrelativelylargeeffectsize.However,identifyingCNVs
fromWEShasbeenachallengebecauseofthesparsenessofthetargetregionsandthenon‐uniform
distributionofreadsacrossgenome.AspartoftheCMGproject,weappliedfourprevailingCNV
callingmethods(XHMM,CoNIFER,ExomeDepth,andEXCAVATOR)on677WESsamples(including
41HapMapcontrols)tosearchforrareexonicCNVsthatmightbecausalforthediseaseofinterest.
Inourpreliminaryanalysis,CoNIFER,ExomeDepth,XHMM,andEXCAVATORdetectedanaverageof
3.5,208,13.3,and58CNVs(forsizeslargerthan300bp)persample,respectively.Ourinitial
analysesofthreeunsolvedconsanguineouspedigreeswiththesamephenotyperevealeda
homozygoustwoexondeletion(~2.45kb)inaknowncausalgeneintwoofthefamilies.Wewill
comparetheresultsbetweenmethods,examinetheimpactofcontrolsused,andreviewasubsetof
findingsinIGV.
Categories: Association:Family‐based,Association:Genome‐wide,Bioinformatics,Case‐ControlStudies,
Causation,CopyNumberVariation,DataIntegration,DataMining,FineMapping,GenomicVariation,
LinkageandAssociation,SequencingData
P26
Combininggeneticandepigeneticinformationidentifiedimprinted4q35
variantassociatedwiththecombinedasthma‐plus‐rhinitisphenotype
ChloéSarnowski1,CatherineLaprise2,MiriamMoffatt3,GiovanniMalerba4,AndréanneMorin2,
QuentinVincent5,KlausRohde6,Marie‐HélèneDizier1,JorgeEsparza‐Gordillo6,Emmanuelle
Bouzigon1
11)U946,INSERM,PARIS,France;2)UniversitéParisDiderot,SorbonneParisCité,InstitutUniversitaire
d’Hématologie,France
23)UniversitéduQuábecàChicoutimi,Canada
34)NationalHeartLungInstitute,ImperialCollege,UK
45)SectionofBiologyandGenetics,DepartmentofLifeandReproductionSciences,UniversityofVerona,Italy
56)U1163,INSERM,PARIS,France
67)Max‐Delbrück‐CenterforMolecularMedicine(MDC),Berlin,Germany
Wepreviouslydetectedalinkagesignalinthe4q35regionwiththecombinedasthma‐plus‐rhinitis
phenotype(AST+AR)in615Europeanfamilieswhenaccountingformaternalimprinting(p=7x10‐
5).Tofurtherinvestigatethisregion,wetestedtheassociationbetween1,300SNPs(spanning6
Mb)andAST+ARin162FrenchEGEAfamiliesascertainedthroughasthmausingtheParent‐of‐
Origin‐LikelihoodRatioTest.Replicationanalysiswasperformedin152asthmaticFrenchCanadian
SLSJfamiliesfor18SNPsdetectedatp<0.005.Thetop‐replicatedSNP(rs10009104)lyingat1.6Mb
fromthelinkagepeakwasdetectedunderabest‐fittingmaternalimprintingmodel(pmeta=4x10‐5)
andaccountedformostofthelinkagesignal.Manycis‐regulatoryelementsaredescribedina50kb
surroundingregionofthisSNP.UsingtheQuantitativeTransmissionDisequilibriumTest(QTDT),
wetestedforassociationbetweenrs10009104and26DNAmethylationprobesofthatregion,
measuredinwhitebloodcellsof159individuals(40SLSJfamilies),whileaccountingforparent‐of‐
origineffectandadjustingforAST+AR.Maternallyinheritedriskalleleofrs10009104was
associatedwithincreasedmethylationofthetop‐rankedprobe(p<10‐5afterpermutations).This
probeliesat529bpfromtheSNPandwithinregulatoryelementsthatincludeapredictedactive
promoterinlungfibroblasts,DNaseIhypersensitiveclusters,andbindingsitesoftwotranscription
factorsinvolvedininflammatoryresponseinitiation(RelAandNF‐κB).Thisstudyidentifieda
maternallyimprintedSNPthataffectsAST+ARthroughanepigeneticmechanism.Funded:Conseil
RégionalIledeFrance,ANRGWIS‐AM,EC‐FP6
Categories: Association:Family‐based,EpigeneticData,LinkageandAssociation,Multifactorial
Diseases,TransmissionandImprinting
P27
BAYESIANLATENTVARIABLECOLLAPSINGMODELFORDETECTING
RAREVARIANTINTERACTIONEFFECTINTWINSTUDY
LiangHe1,MikkoJSillanpää2,SamuliRipatti3,JannePitkäniemi4
1DepartmentofPublicHealth,HjeltInstitute,UniversityofHelsinki,Finland
2DepartmentofMathematicalSciences,UniversityofOulu,OuluFIN‐90014,Finland;DepartmentofBiology
andBiocenterOulu,UniversityofOulu,OuluFIN‐90014,Finland
3InstituteforMolecularMedicineFinlandFIMM,UniversityofHelsinki,Finland;WellcomeTrustSanger
Institute,UK
4FinnishCancerRegistry,InstituteforStatisticalandEpidemiologicalCancerResearch,Helsinki,Finland;
DepartmentofPublicHealth,HjeltInstitute,UniversityofHelsinki,Finland
Byanalysingmorenext‐generationsequencedatathanbefore,researchershaveaffirmedthatrare
geneticvariantsarewidespreadamongpopulationsandlikelyplayanimportantroleincomplex
phenotypes.Recently,ahandfulofstatisticalmodelshavebeendevelopedtoanalyserarevariant
associationindifferentstudydesigns.However,duetothescarceoccurrenceofminorallelesin
data,appropriatestatisticalmethodsfordetectingrarevariantinteractioneffectsarestilldifficultto
develop.WeproposeahierarchicalBayesianlatentvariablecollapsingmethod(BLVCM),which
circumventstheobstaclesbyparameterizingthesignalsofrarevariantswithlatentvariablesina
Bayesianframeworkandisparameterisedfortwindata.TheBLVCMmanagestotacklenon‐
associatedvariants,allowbothprotectiveanddeleteriouseffects,captureSNP‐SNPsynergistic
effect,provideestimatesforthegenelevelandindividualSNPcontributions,andcanbeappliedto
bothindependentandvarioustwindesigns.WeassessthestatisticalpropertiesoftheBLVCMusing
simulateddata,andfindthatitachievesbetterperformanceintermsofpowerforinteractioneffect
detectioncomparedtotheGranvilandtheSKAT.Asproofofpracticalapplication,theBLVCMis
thenappliedtoatwinstudyanalysisofmorethan20,000generegionstoidentifysignificantrare
variantsassociatedwithlow‐densitylipoproteincholesterol(LDL‐C)level.Theresultsshowthat
someofthefindingsareconsistentwithotherpreviousstudies,andsomenovelgeneregionswith
significantSNP‐SNPsynergisticeffectsareidentified.Keywords:rarevariant;Bayesiancollapsing
model;geneticassociation;LDL‐C;twinstudy
Categories: Association:Family‐based,Association:Genome‐wide,BayesianAnalysis,Gene‐Gene
Interaction,GenomicVariation,MarkovChainMonteCarloMethods,MultilocusAnalysis,Pathways,
PopulationGenetics,QuantitativeTraitAnalysis,SequencingData
P28
RareVariantAssociationTestforNuclearFamilies
Zong‐XiaoHe1,NiklasKrumm2,GaoTWang1,BrianJO'Roak3,SimonsSimplexSequencing
Consortium,EvanEEichler3,SuzanneMLeal3
1CenterforStatisticalGenetics,DepartmentofMolecularandHumanGenetics,BaylorCollegeofMedicine
2DepartmentofGenomeSciences,UniversityofWashington
3DepartmentofMolecularandMedicalGenetics,OregonHealthandScienceUniversity
Population‐basedcomplextraitassociationstudiesofrarevariants(RVs)arevulnerabletospurious
associationsduetopopulationstratification.AnalyzingtriodatausingtheRV‐transmission
disequilibriumtest[RV‐TDT(Heetal.2014)]canavoidthisproblem.TheTDTanalysesonlyemploy
informationonanaffectedoffspringandtheirparents.Whentherearesiblings,includingthemin
analysiscanprovideadditionalassociationinformation.WeextendedtheRV‐TDTtoanalyzeall
typesofindependentnuclearfamilies(NF)withatleastoneaffectedoffspring(RV‐NF).ForallRV‐
NFteststypeIerroriswellcontrolledevenwhenthereisahighlevelofpopulationstratificationor
admixture.ThepoweroftheRV‐NFtestwasevaluatedusinganumberofdiseasemodelsand
nuclearpedigreeconfigurations.TheRV‐NFisconsiderablymorepowerfulthantheRV‐TDTto
detectassociations.FortheRV‐TDTandRV‐NFpowerwasevaluatedbygeneratingdatafora
1,500bpgeneforwhichthecausalRVshaveanoddsratioof2.Thepowertodetectandassociation
is:0.49for1,000trios;0.58for1,000NFwithoneaffectedchildandanunaffectedchild;and0.65
for1,000NFwithtwoaffectedchildren.InordertoillustratetheapplicationoftheRV‐NFmethods,
theexomedatafrom600autismspectrumdisorderNFwithoneaffectedchildandoneunaffected
childwereanalyzed.RVassociationswithautismwerefoundforseveralgenes.Giventheproblem
ofadequatelycontrollingforpopulationstratificationandadmixtureinRVassociationstudies,the
capabilityofanalyzingalltypesofNFsandthegrowingnumberofNFstudieswithsequencedata,
theRV‐NFmethodisextremelybeneficialtoelucidatetheinvolvementofRVsindiseaseetiology.
Categories: Association:Family‐based,SequencingData
P29
Samplesizeandpowerdeterminationforassociationtestsincase‐parent
triostudies
HolgerSchwender1,ChristophNeumann2,MargaretATaub3,SamuelGYounkin4,TerriHBeaty3,
IngoRuczinski3
1HeinrichHeineUniversity
2TUDortmundUniversity
3JohnsHopkinsUniversity
4UniversityofWisconsin
Transmission/disequilibriumtests(TDTs)arethemostpopularstatisticaltestsfordetectingsingle
nucleotidepolymorphisms(SNPs)associatedwithdiseaseincase‐parenttriostudiesconsidering
genotypedatafromchildrenaffectedbyadiseaseandfromtheirparents.Sinceseveraltypesof
theseTDTshavebeendevised,e.g.,approachesbasedonallelesorongenotypes,itisofinterestto
evaluatewhichoftheseTDTshavethehighestpowerinthedetectionofSNPsassociatedwith
disease.SincetheteststatisticofthegenotypicTDT–whichisequivalenttoaWaldtestina
conditionallogisticregressionmodel–hadtobecomputednumerically,comparisonsofotherTDTs
withthegenotypicTDThavesofarbeenbasedonsimulationstudies.Recently,we,however,have
derivedaclosed‐formsolutionforthegenotypicTDTsothatthisanalyticsolutioncanbeusedto
deriveequationsforpowerandsamplesizecalculationforthegenotypicTDT.Inthispresentation,
weshowhowtheseequationscanbederivedandcomparethepowerofthegenotypicTDTwiththe
oneofthecorrespondingscoretestassumingthesameunderlyinggeneticmodeofinheritanceas
wellastheallelicTDTbasedonamultiplicativemodeofinheritance.
Categories: Association:Family‐based,Association:Genome‐wide,LinkageandAssociation,Maximum
LikelihoodMethods,SampleSizeandPower
P30
IntegrationofDNAsequencevariationandfunctionalgenomicsdatato
infercausalvariantsunderlyingchemotherapeuticinducedcytotoxicity
response
RuowangLi1,DokyoonKim1,ScottMDudek1,MarylynDRitchie1
1CenterforSystemsGenomics,ThePennsylvaniaStateUniversity,StateCollege,PA
Carboplatinisawidelyusedchemotherapeuticdrugforovarianandlungcancer.Despiteitsbroad
usage,somepatientsexperienceseveresideeffectsincludingmyelosuppressionandmucositis.
Understandingthedrug‐inducedcytotoxicitycouldpotentiallyleadtopersonalizedtreatment.
However,findingthecausalgeneticvariantsthatinfluencethedrug’scytotoxicityhasbeen
challenging.Toidentifyvariantsthatarekeyforcarboplatinresponse,weperformedananalysis
thatjointlyanalyzedDNAsequencevariationandfunctionalgenomicsdatainCEUandYRIHapMap
populations.CarboplatinresponsewasmeasuredontheCEUandYRIlmphoblastoidcelllinesin
termsofIC50,concentrationrequiredtostop50%ofcellgrowth.Usingwholegenomesequencing
datafromthe1000GenomesProjectandRNAsequencingdatafromtheGEUVADISproject,we
identifiedcandidategeneticvariantsandgeneexpressionvariablesthatareassociatedwith
carboplatinIC50.Touncoverpotentialinteractionsbetweencandidatevariantsandgene
expressionfactors,weintegratedthecandidatesusinggrammaticalevolutionneuralnetwork
implementedinATHENA.Theintegrationanalysisidentifieduniquesetsofgeneticvariantsand
geneexpressionfactorsininteractionmodelsinbothCEUandYRIpopulationwithhighpredictive
power(R2>60%).Toavoidselectionbias,wealsoidentifiedvariantsthatareinlinkage
disequilibriumwiththecandidatevariants.Wethenprioritizedallthevariantsbasedonhundreds
offunctionalgenomicannotationsfromtheENCODEproject,includinggenes,enhancers,and
DNase‐Isites.Basedontheconsistencyandenrichmentoffunctionalannotations,wefound
potentialcausalvariantsforcarboplatinresponse.
Categories: Association:Genome‐wide,Cancer,Causation,DataIntegration,GenomicVariation
P32
ImputationforSNPsusingsummarystatisticsandcorrelationbetween
genotypedata
SinaRüeger1,2,ZoltánKutalik1,2
1InstituteofSocialandPreventiveMedicine,UniversityHospitalandUniversityofLausanne,Lausanne
2SwitzerlandSwissInstituteofBioinformatics,Lausanne,Switzerland
Genome‐wideassociationstudiesusemicroarraystomeasureSNPsthatareoftendesignedtotag
manyuntypedvariants,whichcanbeimputedviathelinkagedisequilibrium(LD)between
measuredanduntypedmarkers.Theimputationmethods,whilemakingmostoftheavailabledata,
arecomputationallyveryexpensivewhenitcomestoimputing~30‐40Mvariantsofthe1000
Genomespanel.Theseimputedvariantsaresubsequentlysubjectedtoassociationwithvarious
traits.
Weproposeanapproachthatperformsimputationdirectlyontheassociationsummarystatistics
(suchast‐statistics)oftypedSNPs.Thisallowsafastinferenceoftheassociationstrengthofnon‐
genotypedmarkersusingthatofthetaggingSNPs.Thisapproachbearssimilaritieswiththe
pioneeringworkofPasaniucetal.(2013).Thenoveltyofourmethodliesintheoptimized
regularizationofthepair‐wisemarkercorrelationmatrix,amodifiedconditionalexpectation.Italso
allowsforassociationsderivedfromdifferentsamplesizes.Wereachedfurtherimprovementsby
selectingthemostrelevantreferencehaplotypesetsinordertoimputesummarystatistics.
Fortestingweusedthelipidassociationmeta‐analysessummarystatisticsfromWilleretal.(2013).
UsingtheassociationstatisticsfromHapMapSNPsonly,weimputedtheeffectsizeofnon‐HapMap
SNPsandcomparedtothe“true”effectsizeestimatesresultingfromgenotypeimputationand
association.Theresultssuggestthatourteststatisticsagreecloser(r2=0.87)withthetruevalues
thantheestimatesprovidedbypreviousmethods(r2=0.82).
Suchfastandaccurateimputationmethodswillbecomeincreasinglyimportantasreferencepanels
growinsizeandgenotypeimputationturnsouttobelessfeasible.
PasaniucB,ZaitlenN,ShiH,BhatiaG,GusevA,PickrellJ,HirschhornJ,StrachanDP,PattersonN,
PriceAL(2013)Fastandaccurateimputationofsummarystatisticsenhancesevidenceoffunctional
enrichment.ArXiv:1309.3258v1[q‐bio.QM]
WillerCJ,SchmidtEM,SenguptaS,PelosoGM,GustafssonS,KanoniS,GannaA,ChenJ,Buchkovich
ML,etal.;Globallipidsgeneticsconsortium(2013)Discoveryandrefinementoflociassociatedwith
lipidlevels.NatureGenetics45(11),1274‐1283.
Categories: Association:Genome‐wide,LinkageandAssociation,MissingData
P33
Evaluationofpopulationstratificationinalargebiobanklinkedto
ElectronicHealthRecords
MarizadeAndrade1,GerardThromp2,AmberBurt3,DanielSKim4,ShefaliSVerma3,AnastasiaM
Lucas3,SebastianMArmasu1,JohnA.Heit1,GeoffreyMHayes5,HelenaKuivaniemi2
1MayoClinic,Rochester,MN,USA
3PennsylvaniaStateUniversity,UniversityPark,PA,USA
4UniversityofWashington,Seattle,WA,USA
5NorthwesternUniversity,Chicago,IL,USA
Forgenomicassociationstudies,combiningsamplesacrossmultiplestudiesinNetworksor“Big
Science”isstandardpractice.Increasingthenumberofsubjectsallowsforpowerneededtoassess
association.Controllingforgenomicancestryiscommon,butthereisaneedtostandardizethe
approachwhencalculatingprincipalcomponents(PCs)acrosscohortssuchaseliminationofSNPs
withlinkagedisequilibrium(LD)pruningatr=0.5andaMAF<0.03.Duetoheterogeneitybetween
sites,adjustingforPCsonly,doesnotremovethesiteandplatformbias.Therefore,weproposean
alternativeapproachofgeneratingPCsforourcohorttocontrolforsiteandplatformbiasin
additiontoancestrydifference.OurapproachconsistsonderivingthePCsusingtheloadings
calculatedfromreferencesamples,muchlikegeneratingPCsusingthefoundersoffamilies.We
appliedourapproachusingtheelectronicMedicalRecordsandGenomics(eMERGE)Venous
ThromboembolismAfricanancestrycohortthatconsistsoffouradultsitesandfourgenotyping
platformsthathadpreviouslybeenanalysedcontrollingforsite,platformandancestry.Ourresults
showedthatourapproachprovidedsimilarassociationresultswhilebothcontrollingforinflation(λ
=1.01and1.02forstandardandloadings,respectively)withtheadvantagesofcontrollingforfewer
covariates,thuslessdegreesoffreedom.Therefore,weexpectthisapproachwillserveasa“Best
Practices”forsimilarprojects,andasareferenceforassessingandcontrollingforconfoundersin
additiontoancestryingeneticassociationstudies.
Categories: Association:Genome‐wide,PopulationStratification
P34
Estimatinggeneticeffectsonsusceptibilityandinfectivityforinfectious
diseases
FloorBiemans1,PiterBijma2,MartCMDeJong3
1QuantitativeVeterinaryEpidemiologyGroup,WageningenUniversity;AnimalBreedingandGenomicsCentre,
WageningenUniversity
2AnimalBreedingandGenomicsCentre,WageningenUniversity
3QuantitativeVeterinaryEpidemiologyGroup,WageningenUniversity
Transmissionofinfectiousdiseasesisdeterminedbysusceptibilityandinfectivityoftheindividuals
involved.Anindividual’sgenesforsusceptibilityaffectthediseasestatusoftheindividualitself,and
thusrepresentadirectgeneticeffect.Anindividual’sgenesforinfectivity,ontheotherhand,affect
thediseasestatusofotherindividuals,andthusrepresentaso‐calledindirectgeneticeffect(IGE).
AnIGEisageneticeffectofanindividualonthephenotypeofanotherindividual.IGEshavebeen
studiedextensivelyinevolutionarybiology,andcanhaveprofoundeffectsontherateanddirection
ofevolutionbynaturalselection.Ingeneticstudiesoninfectiousdiseases,thecurrentfocusis
largelyonsusceptibility,whereasgeneticsofinfectivitycanhavemajoreffectsondisease
transmission.However,littleisknownaboutthegeneticbackgroundofinfectivity.Weshowhow
geneticeffectsonsusceptibilityandinfectivitycanbeestimatedsimultaneouslyfromtime‐series
dataondiseasestatusofindividuals.Anendemicdiseasewassimulated,andthediseasestatus
(0/1)andgenotypeofindividualswererecordedatseveralpointsintime.Thesedatawere
analysedusingageneralizedlinearmodel(GLM)withacomplementarylog‐loglinkfunction.The
modelincludedtwogeneticterms:i)thegenotypeofthefocalindividual,representing
susceptibility,andii)theaveragegenotypeofitsinfectedsocialpartners(contacts),representing
infectivity.Firstresultsshowedthatestimatedgeneticeffectswerealmostunbiased.Thiswork,
therefore,providesatoolforgenome‐wideassociationstudiesaimingtoidentifygenomicregions
affectingsusceptibilityandinfectivityofindividualstoendemicdiseases.
Categories: Association:Genome‐wide,GenomicVariation,HaplotypeAnalysis,PredictionModelling
P35
CombinedMethodstoExploreGeneticEtiologyofRelatedComplex
Diseases
ShefaliSetiaVerma1,AnuragVerma1,AnastasiaLucas1,JimLinneman2,PeggyPeissig2,Murray
Brilliant2,CatherineAMcCarty3,JonathanLHaines4,TamaraRVrabec5,GerardTromp5
1CenterforSystemsGenomics,DepartmentofBiochemistryandMolecularBiology,PennsylvaniaState
University,UniversityPark,PA,USA
2MarshfieldClinic,Marshfield,WI,USA
3EssentiaRuralHealth,Duluth,MN,USA
4CaseWesternUniversity,Cleveland,OH,USA
Genome‐wideassociationstudies(GWAS)haveidentifiedseveralSNPsassociatedwitheither
glaucomaorocularhypertension(OHT).However,thesesusceptibilitylociexplainasmallfraction
ofthegeneticrisk.Gene‐geneinteraction(GxG)studiesareconsideredapotentialavenuetoidentify
thismissingheritability.UsingadatasetfromtheeMERGE(electronicMedicalRecordsand
Genomics)Network,whichincludedGWASdataimputedusingthe1000Genomes,wewereableto
explorethegeneticetiologyoftwoveryrelatedcommoneye‐diseases:glaucomaandOHT.OHTis
oneoftheleadingriskfactorforglaucoma,thusweexploredtherelationshipsbetweenthesetwo
traitsatthemolecularlevel.Atotalof3,253(glaucoma)and3,154(OHT)unrelatedsamplesofages
40‐90wereextractedfromtheeMERGEstudybiorepositories.First,weperformedGWASandGxG
studiesforeachtraitusingtheimputeddatasetandidentifiedseveralmaineffectsandGxGmodels
thatmeetBonferronisignificance.Secondly,fromtheobtainedGWASwithmaineffectp<0.01,we
alsoperformedapathway‐enrichmentanalysisusingKEGGdatabaseonbothofthesetraits
combined.Interestingly,weobservedthatgenesinABCtransporterpathwayarefoundtobe
associatedwithbothglaucomaandOHT.TheABCA4geneishighlyassociatedwithglaucomaand
alsoshowssignificantinteractionwithGAD2geneinOHT(p=2.71x10‐11).Lastly,outof10
pathwayssharedbetweenthetwotraits,ABCtransportergenesarefoundtobehighlyassociated
withboththetraits.Inconclusion,wewereabletoidentifynovelSNPassociationsandGxG
interactionsforthesetraitsanddemonstratetherelationshipbetweenthesetwotraitsatthe
molecularlevelwiththeguidanceofpathwayanalysis.
Categories: Association:Genome‐wide,Association:UnrelatedCases‐Controls,Bioinformatics,Gene‐
GeneInteraction,Pathways
P36
Integrativeanalysisofsequencingandarraygenotypedatafor
discoveringdiseaseassociationswithraremutations
YijuanHu1,YunLi2,PaulLAuer3,DanyuLin4
1DepartmentofBiostatisticsandBioinformatics,EmoryUniversity,USA
2DepartmentofBiostatistics,DepartmentofGenetics,UniversityofNorthCarolina,ChapelHill,USA
3JosephJ.ZilberSchoolofPublicHealth,UniversityofWisconsin,Milwaukee,USA
4DepartmentofBiostatistics,UniversityofNorthCarolina,ChapelHill,USA
Inthelargecohortstypicallyusedforgenome‐wideassociationstudies(GWAS),itisprohibitively
expensivetosequenceallcohortmembers.Acost‐effectivestrategyistosequencesubjectswith
extremevaluesofquantitativetraitsorthosewithspecificdiseases.Byimputingthesequencing
datafromtheGWASdataforthecohortmemberswhoarenotselectedforsequencing,onecan
dramaticallyincreasethenumberofsubjectswithinformationonrarevariants.However,treating
theimputedrarevariantsasobservedquantitiesindownstreamassociationanalysismayinflatethe
typeIerror,especiallywhenthesequencedsubjectsarenotarandomsubsetofthewholecohort.
Althoughtheproblemcanbealleviatedbyrestrictingtheanalysistovariantsthatareaccurately
imputed,alargenumberofrarevariantswillbeexcludedasaresult.Inthisarticle,weprovidea
validandefficientapproachtocombiningobservedandimputeddataonrarevariants.Weconsider
allcommonlyusedgene‐levelassociationtests,includingtheburdentest,variablethreshold(VT)
test,andsequence‐kernelassociationtest(SKAT),allofwhicharebasedonthescorestatisticfor
assessingtheeffectsofindividualvariantsonthetraitofinterest.Weshowthatthescorestatistic
basedontheobservedgenotypesforsequencedsubjectsandtheimputedgenotypesfornon‐
sequencedsubjectsisunbiased.Weconstructarobustvarianceestimatorthatreflectsthetrue
variabilityofthescorestatisticregardlessofthesamplingschemeandimputationquality,suchthat
thecorrespondingassociationtestsalwayshavecorrecttypeIerror.Wedemonstratethrough
extensivesimulationstudiesthattheproposedtestsaresubstantiallymorepowerfulthantheuseof
accuratelyimputedvariantsonlyandtheuseofsequencingdataalone.Weprovideanapplicationto
theWomen'sHealthInitiative(WHI).Therelevantsoftwareisfreelyavailable.
Categories: Association:Genome‐wide,DataIntegration
P37
Amethodforfastcomputationoftheproportionofvariantsaffectinga
complexdiseaseandoftheadditivegeneticvarianceexplainedinGWAS
SNPstudies.
LuigiPalla1,FrankDudbridge1
1DepartmentofNon‐communicableDiseaseEpidemiology,LondonSchoolofHygieneandTropicalMedicine
RecentresearchhasaddressedtheestimationofvarianceexplainedbylargesetsofSNPsfroma
genomewidepanel.AmethodbasedonpolygenicscoringwasproposedbyStahletal(NatGenet
2012)toestimatebothvarianceexplainedandnumberofSNPsaffectingthetrait,via
computationallyintensiveBayesianmethodology.Weproposeafastanalyticmethodbasedonthe
formulaforthenoncentralityparameteroftheassociationtestofapolygenicscorewiththetraitof
interest(Dudbridge,PLoSGenet2013).Weshowhowmodelparameterscanbeestimatedfromthe
resultsofmultiplepolygenicscoretestsbasedonSNPswithP‐valuesfallingindifferentintervals.
Weestimatemodelparametersusingmaximumlikelihoodanduseaprofilelikelihoodapproach
thatallowsrapidcomputationofreliableconfidenceintervals.Weillustrateourmethodonseveral
examplesofcomplexdiseases.Wecomparevariouschoicesforconstructingpolygenicscores,based
onnestedordisjointintervalsofp‐valuesandonweightedorunweightedSNPeffectsizes,in
estimatingvarianceexplained(vg),fractionofgenesaffectingthetrait(nf)andcovariancebetween
effectsintrainingandreplicationsamples.Wefindthatforestimationofvgandnfonly,the
estimatesarenearlyunbiasedandconfidenceintervalsnarrow,withlessbiasfordisjointintervals.
Whenestimatingall3parameterstheestimatespresentevensmallerbias,largerconfidence
intervals,butincuralargerbiasforvginthecaseofnestedintervals.Overallwerecommenduseof
thismethodbasedontheresultsderivedfromdisjointintervals.
Categories: Association:Genome‐wide,Case‐ControlStudies,MaximumLikelihoodMethods,
QuantitativeTraitAnalysis
P38
Correctingforsampleoverlapincross‐traitanalysisofGWAS
MarissaLeBlanc1,VerenaZuber2,ArnoldoFrigessi3,BettinaKulleAndreassen1
1Epi‐Gen,InstituteofClinicalMedicine,AkershusUniversityHospital,UniversityofOslo,Oslo,Norwayand
OsloCentreforBiostatisticsandEpidemiology,DepartmentofBiostatistics,UniversityofOslo,Norway
2NORMENT,KGJebsenCentreforPsychosisResearch,InstituteofClinicalMedicine,UniversityofOslo,Oslo,
Norway,DivisionofMentalHealthandAddiction,OsloUniversityHospital,Oslo,NorwayandProstateCancer
ResearchGroup,CentreforMolecularMe
3OsloCentreforBiostatisticsandEpidemiology,DepartmentofBiostatistics,UniversityofOslo,Norwayand
StatisticsforInnovation,NorwegianComputingCenter,Oslo,Norway
Thereisagrowinginterestinintegratinggenomicdataoverdifferenttraits,atthesummary
statisticslevel.Thisisofbiologicalinterestduetothepartiallysharedgeneticbasisofmanytraits,
termedpleiotropy.Using,forexample,meta‐analysisoraconditionalfalsediscoveryrate(FDR)
framework,pleiotropycanbeleveragedtoimprovedetectionofcommongeneticvariantsinvolved
indisease.Thisrequiresonlysummarystatistics,notindividual‐leveldata.Summarystatistics
fromgenome‐wideassociationstudies(GWAS)conductedbyglobalconsortiaarebecomingeasier
toobtain,howevertheseGWASsummarystatisticsareoftennotindependentacrosstraitsdueto
partiallyoverlappingsamples.Ouraimsaretwofold.First,weshowtheimpactofsampleoverlap
oncross‐traitanalysisofGWAS,anddemonstratewithsimulationsthatitcaninducespurious
correlationandanincreasedproportionoffalsepositivefindings.Second,weproposeacorrection
thatremovesthespuriouseffectsduetosampleoverlap.Thiscorrectioninvolvesfirstestimating
thecorrelationofthesummarystatisticsfromthetwostudies(forallpossiblecombinationsof
quantitativeandbinaryoutcomes),andthensecond,correctingforthisspuriouscorrelationviathe
Mahalanobistransformation.WepresentresultsfromsimulationstudiesandfromactualGWAS
datathatshowthattheproposedcorrectionforsampleoverlapproperlycontrolsforfalsepositive
findingswhilestillallowingforthedetectionoftruepleiotropicfindings.
Categories: Association:Genome‐wide,DataIntegration
P39
Epigenome‐wideassociationstudyofcentralizedadiposityin2,083
AfricanAmericans:TheAtherosclerosisRiskinCommunities(ARIC)
Study
LindsayFernández‐Rhodes1,YunLi1,MariaelisaGraff1,WeihuaGuan2,MeganLGrove3,QingDuan1,
GuoshengZhang1,MyriamFornage3,JamesPankow2,EllenWDemearath2
1UniversityofNorthCarolinaatChapelHill,NorthCarolina,USA
2UniversityofMinnesota,Minnesota,USA
3UniversityofTexasHealthScienceCenteratHouston,Texas,USA
CentralobesityisaleadingpredictorofcardiometabolicriskanditsprevalenceintheUnitedStates
(US)hasmorethandoubledsincethe1980s,especiallyinUSminorities.Evidencesuggeststhat
geneticfactorscontributetocentraladiposity,measuredaswaisttohipratioadjustedforbody
massindex(WHRa).DNAmethylationpatterns,awell‐studiedformofepigeneticmodiﬁcation,may
alsoassociatewithWHRa.Thisstudyaimstoexaminethecross‐sectionalassociationbetween
genome‐wideCpGsitemethylationandWHRainAfricanAmericans.
TheInfiniumHumanMethylation450KBeadChipwasusedtomeasuremethylationinbisulphite‐
convertedperipheralbloodDNAfrom2,083AfricanAmericans(meanage56.6years)inthe
AtherosclerosisRiskinCommunitiesstudy.Linearmixedeffectsmodelswereusedtotestfor
associationbetweenmethylationbetavaluesandWHRaaccountingforrandomeffectsforbatchand
fixedeffectsforage,sex,center,education,concurrentwhitebloodcellcount,householdincome,
smoking,alcoholconsumption,physicalactivity,fiveleukocytecelltypeproportions,andprincipal
componentsderivedfromgenome‐wideexonicgenotypedata.
WeobservedonesignificantnegativeassociationwithWHRaatcg00574958(p=7x10‐12),which
liesinthe5'UTRofCPT1A,agenepreviouslyimplicatedwithmetabolicrelatedtraits.Weaker
associations(p<1x10‐6)werealsoobservedatseveralautosomalsitesrequiringfuture
independentreplication.
OurobservedCpGsiteassociationataknownmetaboliclocussuggeststhatepigeneticsignaturesof
centraladipositymayaccountforsomeofthemissingheritabilityandinformourunderstandingof
metabolicdysregulation.
Categories: Association:Genome‐wide,CardiovascularDiseaseandHypertension,EpigeneticData,
Epigenetics
P40
Canlow‐frequencyvariantsberescuedingenome‐wideassociation
studiesusingsparsedatamethods?
Ji‐HyungShin1,ShelleyBBull1
1Lunenfeld‐TanenbaumResearchInstituteofMountSinaiHospital&DallaLanaSchoolofPublicHealth,
UniversityofToronto
Formanycomplextraits,geneticvariantsthatoccurwithlowfrequency(MAF<5%)arethoughtto
beimportant.However,genome‐widescansofbinarytraitsusuallyexcludelowfrequencyvariants
evenwhenthesamplesizeismoderate,becauseconventionallogisticregressioninferencecanfail
duetolowcountsoftheobservedlow‐frequencyvariants.Alternatively,sparsedatamethodssuch
asFirth’spenalizedlogisticregressionlikelihoodratiotestorasmall‐sample‐adjustedscoretest
implementedinSKATcanprovidevalidresultsforasinglevariant.Weinvestigatetheperformance
ofthestandardlogisticregressionandthesparsedatamethods,usinganalyticderivationsand
finite‐samplesimulationsacrossvariousscenarios.Intheanalyticinvestigation,weexaminethe
simplecaseofa2‐by‐2contingencytabletogaininsightintodifferencesamongthemethods.The
analyticcalculationsshowhowtheteststatisticsdependontheobservednumbersofaffectedand
unaffectedindividualswiththelow‐frequencyvariant,andonthediseaseprevalence.Inthe
simulationstudy,weconsideranadditivelycodedgenotypeandaquantitativecovariate,andvary
diseaseprevalence,minorallelefrequencyandcounts,andeffectsizeofthegeneticcovariateto
examinearangeofsettings.Wefindthatnoonetestisuniformlybetterthantheothers.Overall,
type1errorratesareclosesttothenominallevelforthepenalizedlikelihoodratiotestandthe
small‐sample‐adjustedscoretest,whiletype1errorratesfortheothertestscanbegreatlyinflated
ordeflated.Thepowerforthesmall‐sample‐adjustedscoretesttendstobeslightlyhigherthanthe
penalizedlikelihoodratiotest,butthedifferencemaybeinsignificantinpractice.
Categories: Association:Genome‐wide,MaximumLikelihoodMethods
P41
Anovelkernel‐basedstatisticalapproachtotestingassociationin
longitudinalgeneticstudieswithanapplicationofalcoholusedisorder
inaveterancohort
ZuohengWang1,ZhongWang1,JosephL.Goulet1,JohnH.Krystal1,AmyC.Justice1,KeXu1
1YaleUniversity
Alcoholdependence(AD)isamajorpublichealthconcernintheUnitedStatesandcontributesto
thepathogenesisofmanydiseases.TheriskforADismultifactorialincludingbothgeneticand
environmentalfactors.Currently,theconfirmedassociationsaccountforasmallproportionof
overallgeneticrisksforAD.Multiplemeasurementsinlongitudinalgeneticstudiesprovidearoute
toreducenoiseandcorrespondinglyincreasethestrengthofsignalsingenome‐wideassociation
studies(GWAS).Inthisstudy,wedevelopedapowerfulkernel‐basedstatisticalmethodfortesting
thejointeffectofgenevariantswithageneregionondiseaseoutcomesmeasuredovermultiple
timepoints.Weappliedthenewmethodtoalongitudinalstudyofveterancohort(N=960)with
bothHIV‐infectedandHIV‐uninfectedpatientstounderstandthegeneticriskunderlyingAD.We
foundaninterestinggenethatmayinvolvetheinteractionofHIVreplication,suggestiveofpotential
genebyenvironmenteffectinalcoholuseandHIV.Wealsoconductedsimulationstudiestoaccess
theperformanceofthenewstatisticalmethodsanddemonstratedapowergainbytakingadvantage
ofrepeatedmeasurementsandaggregatinginformationacrossabiologicalregion.Thisstudynot
onlycontributestothestatisticaltoolboxinthecurrentGWAS,butalsopotentiallyadvancesour
understandingoftheetiologyofAD.Acknowledgment:TheauthorsthanktheVeteransAging
CohortStudyandVANationalCenterforPTSDforgeneroussupport.ThestudyissupportedbyNIH
grantR21AA022870.
Categories: Association:Genome‐wide,MultipleMarkerDisequilibriumAnalysis,PsychiatricDiseases
P42
AGene‐EnvironmentInteractionBetweenCopyNumberBurdenand
OzoneExposureinRelationtoRiskofAutism
DokyoonKim1,HeatherVolk2,SarahAPendergrass1,MollyAHall1,ShefaliSVerma1,Santhosh
Girirajan1,IrvaHertz‐Picciotto3,MarylynDRitchie1,ScottBSelleck1
1DepartmentofBiochemistry&MolecularBiology,thePennsylvaniaStateUniversity,UniversityPark,PA
2DepartmentofPreventiveMedicine,KeckSchoolofMedicine,UniversityofSouthernCalifornia,LosAngeles,
CA;DepartmentofPediatrics,Childrenâ€™sHospitalLosAngeles,UniversityofSouthernCalifornia,Los
Angeles,CA
3DepartmentofPublicHealthSciences,UniversityofCalifornia,Davis,Davis,CA
Autismisaneurodevelopmentaldisordercharacterizedasacomplextraitwithahighdegreeof
heritabilityaswellasadocumentedsusceptibilityfromenvironmentalfactors.Therelative
contributionsofgeneticfactors,environmentalfactorsandtheirinteractionsastheyrelatetoriskof
autismarepoorlyunderstood.Whilemostautismrelatedcopynumbervariations(CNV)identified
todate,eachwithasubstantialrisk,arehighlypenetrantforthisdisorder,theyconstitutelargerare
eventscontributingmodestlytotheoverallheritability.Genome‐wideanalysisofCNVshave
demonstratedacontinuousriskofautismassociatedwiththegloballevelofcopynumberburden,
measuredastotalbasepairsofduplicationordeletionacrossthegenome.Inaddition,
environmentalexposuretoairpollutantshasbeenidentifiedasariskfactorfordeveloping
autism.WehaveexaminedtherelativecontributionofCNV(measuredastotalbasepairsofcopy
numberburden),exposuretoairpollution,andtheinteractionbetweenairpollutantlevelsand
copynumberburdeninapopulationbasedcase‐controlstudy,ChildhoodAutismRisksfrom
GeneticsandEnvironment(CHARGE).Asignificantandsizableinteractionwasidentifiedbetween
duplicationburdenandozoneexposure(OR2.78,P<0.005),greaterthanthemaineffectforeither
copynumberduplication(OR2.41,95%CI:1.36~4.82)orozonealone(OR1.19,95%CI:
0.75~1.89).Theoverallimplicationofourfindingisthatsignificantgene‐environmentinteractions
associatedwithautismexistandcouldaccountforaconsiderablelevelofheritabilitynotdetected
byevaluatingDNAvariationorenvironmentalone.
Categories: Association:Genome‐wide,CopyNumberVariation,Gene‐EnvironmentInteraction
P43
Choosingacase‐controlassociationteststatisticforlow‐countvariants
intheUKBiobankLungExomeVariantEvaluationStudy
NickShrine1,LouiseVWain1,IoannaNtalla1,JamesPCook1,AndrewPMorris2,EleftheriaZeggini3,
JonathanMarchini4,DavidPStrachan5,IanPHall6,MartinDTobin1
1DepartmentofHealthSciences,UniversityofLeicester,Leicester,UnitedKingdom
2DepartmentofBiostatistics,UniversityofLiverpool,Liverpool,UnitedKingdom
3WellcomeTrustSangerInstitute,Hinxton,Cambridgeshire,UnitedKingdom
4DepartmentofStatistics,UniversityofOxford,Oxford,UnitedKingdom
5PopulationHealthResearchInstitute,StGeorge'sUniversityofLondon,London,UnitedKingdom
6DivisionofTherapeuticsandMolecularMedicine,UniversityofNottingham,Nottingham,UnitedKingdom
TheUKBiobankLungExomeVariantEvaluation(UKBiLEVE)studyisanestedcase‐controlstudyto
evaluategeneticsusceptibilitytochronicobstructivepulmonarydisease(COPD),geneticvariants
associatedwithlungfunctionandgeneticresistancetotobaccosmoke.50KUKBiobankindividuals
weresampledfromtheextremesandmiddleofthelungfunctiondistributioninsmokingandnon‐
smokingstrata.Inordertoidentifyrare,putativefunctionalgeneticvariants,genome‐wide
genotypingwasundertakenusingacustomdesignedAffymetrixarraythatincluded130Krare
missenseandlossoffunctionvariants,642Kvariantsselectedforoptimalimputationofcommon
variationandimprovedimputationoflowfrequencyvariation(MAF1‐5%)and9000variants
selectedforimprovedcoverageofknownandcandidaterespiratoryregions.Simulationshave
shownthatlogisticregressionwiththeusualWaldteststatisticatlowminorallelecount(MAC)is
highlyconservative.AlternativeteststatisticswithmorepoweratlowerMACscanbeanti‐
conservative,havingmarkedlydifferenttypeIerrorratesdependingontheMACandbalanceof
casesandcontrols.Ofthe782KvariantspassingQCinUKBiLEVE,around57KhaveMAC<20with
11Ksingletons;phenotypiccomparisongroupshavecase‐controlratiosofeitherapproximately1:1
or1:2.Wecompareinflationofteststatistics,numberofassociatedlocidetectedandcomputational
efficiencyofthescoreandFirthtestsforassociationtestingofrarevariantsinbalancedand
unbalancedcase‐controlcomparisonsinUKBiLEVE.ThisresearchhasbeenconductedusingtheUK
BiobankResource.
PopulationGenetics,SampleSizeandPower
P44
SNPCHARACTERISTICSPREDICTREPLICATIONSUCCESSINASSOCIATION
STUDIES
IvanPGorlov1,JasonH.Moore1,OlgaYGorlova1,ChristipherIAmos,TheGeiselSchoolofMedicine,
DartmouthCollege
1TheGeiselSchoolofMedicine,DartmouthCollege
TheonlywaytodistinguishtruefromfalsediscoveriesderivedfromGenomeWideAssociation
Studies(GWAS)isreplication.AnindependentreplicationofaSNP/diseaseassociationsuggests
thattheassociationisreal.SelectingSNPsforreplicationstageisbasedonp‐valesfromthe
discoverystage.Reproducibilityofthetopfindingfromdiscoveryphaseislowmakingidentification
ofpredictorsofSNPreproducibilityisimportant.Weuseddisease‐associatedSNPsfrommorethan
2,000publishedGWASstodevelopamodelofSNPreproducibility.Reproducibilitywasdefinedasa
proportionofsuccessfulreplicationsamongallreplicationattempts.Thestudyreporting
SNP/diseaseassociationforthefirsttimewasconsideredtobediscoveryandallconsequentGWASs
targetingthesamephenotypereplications.Wefoundthat‐Log(P),wherePisap‐valuefromthe
discoverystudy,wasthestrongestpredictoroftheSNPreproducibility.Othersignificantpredictors
includetypeoftheSNP(e.g.missensevsintronicSNPs),minorallelefrequencyandeQTLstatusof
theSNP.FeaturesofthegeneslinkedtotheGWAS‐detectedSNPwerealsoassociatedwiththeSNP
reproducibility.Basedonempiricallydefinedrules,wedevelopedasimplifiedreproducibilityscore
(RS)modeltopredictSNPreproducibility.Both‐Log(P)andRSindependentlypredictedSNP
reproducibilityinamultipleregressionanalysis.Weuseddatafrom2lungcancerGWASstudiesas
wellasrecentlyreporteddisease‐associatedSNPstovalidatethemodel.MinusLog(P)outperforms
RSwhenverytopSNPsareselected,whileRSworksbetterwithrelaxedselectioncriteria.In
conclusion,wedevelopedanempiricalmodelforpredictionoftheSNPreproducibility.Themodel
canbeusedforselectionSNPsforvalidationaswellasforSNPprioritizingtobecausal.
Categories: Association:Genome‐wide,Bioinformatics,Cancer
P45
Data‐DrivenWeightedEncoding:ANovelApproachtoBiallelicMarker
EncodingforEpistaticModels
JohnRWallace1,MollyAHall1,ShefaliSVerma1,KristelvanSteen2,ElenaSGusareva2,JasonH
Moore3,BrendanJKeating4,CatherineAMcCarthy5,SarahAPendergrass1,MarylynDRitchie1
1CenterforSystemsGenomics,ThePennsylvaniaStateUniversity,StateCollege,PA
2DepartmentofElectricalEngineeringandComputerScience,UniversityofLiÃ¨ge,LiÃ¨ge,Belgium
3DepartmentofGenetics,GeiselSchoolofMedicineatDartmouthCollege,Lebanon,NH
4CenterforAppliedGenomics,TheChildren'sHospitalofPhiladelphia,Philadelphia,PA
5EssentiaInstituteofRuralHealth,Duluth,MN
WithGenomeWideAssociationStudies(GWAS),biallelicmarkersaretypicallyencodedusingan
additivemodel,assigningvaluesbythenumberofminoralleleseachindividualpossesses.In
detectingmaineffects,thisencodinghasbeenshowntobeanadequatecompromise;however,
choosingoneencodingmakesanassumptionaboutthebiologicalactionofeverymarkerinthe
dataset,whichcanintroduceartifacts.Thisisparticularlyanissuewheninteractiontermsare
added,astheseartifactscanleadtospuriousresults.Analternativeistheuseofcodominant
encoding,whichmakesnoassumptionaboutthebiologicalactionofamarker,butthenumberof
degreesoffreedomrequiredcandramaticallyreducethepowerandintroducecolinearity,
particularlyforinteractionmodels.
Toaddressthesechallenges,wehavedevelopedanovelandeffectiveapproachforencodingthatis
entirelydatadrivenandrequiresnoassumptionsaboutthebiologicalactionofanyparticular
marker,called“Data‐DrivenWeightedEncoding”(DaDWE)Usingtworeal‐worlddatasets:body‐
massindexdatafrom15,737individualsacrossfivedifferentdiversecohortsandage‐related
cataractdatafrom3,377samples(2,192cases;1,185controls)fromtheMarshfieldClinic,weshow
thatthechoiceofencodingcanhavealargeimpact.Foramodelwithonlymaineffects,weshow
thatourmethodhasidenticalresultscomparedtocodominantencoding,andwheninteraction
termsareintroduced,weshowDaDWEhasadistinctadvantageduetoreduceddegreesoffreedom.
Further,usingsimulationdata,weshowthatDaDWEisrobusttomultipletypesofbiologicalactions
underlyingpotentialpredictivemodels,andisanappropriatechoiceforepistaticmodeldiscovery.
EpigeneticData,Epigenetics,Gene‐GeneInteraction
P46
AOne‐Degree‐of‐FreedomTestforSupra‐MultiplicativityofSNPEffects
ChristineHerold1,VitaliaSchüller1,AlfredoRamirez2,TatsianaVaitsiakhovich3,TimBecker1
1GermanCenterforNeurodegenerativeDiseases(DZNE),Bonn,Germany
2DepartmentofPsychiatryandPsychotherapy,UniversityofBonn,Bonn,Germany;InstituteofHuman
Genetics,UniversityofBonn,Bonn,Germany
3InstituteforMedicalBiometry,InformaticsandEpidemiology,UniversityofBonn,Bonn,Germany
Deviationfrommultiplicativityofgeneticriskfactorsisbiologicallyplausibleandmightexplainwhy
Genome‐wideassociationstudies(GWAS)sofarcouldunravelonlyaportionofdiseaseheritability.
Still,evidenceforSNP‐SNPepistasishasrarelybeenreported,suggestingthat2‐SNPmodelsare
overlysimplistic.Inthiscontext,itwasrecentlyproposedthatthegeneticarchitectureofcomplex
diseasescouldfollowlimitingpathwaymodels.Thesemodelsaredefinedbyacriticalriskallele
loadandimplymultiplehigh‐dimensionalinteractions.Here,wepresentacomputationallyefficient
one‐degree‐of‐freedom"supra‐multiplicativity‐test"(SMT)forSNPsetsofsize2to500thatis
designedtodetectriskalleleswhosejointeffectisfortifiedwhentheyoccurtogetherinthesame
individual.ViaasimulationstudyweshowthatouroriginalSMTispowerfulinthepresenceof
thresholdmodels,evenwhenonlyabout30–45%ofthemodelSNPsareavailable.Wecanalso
demonstratethattheSMToutperformsstandardinteractionanalysisunderrecessivemodels
involvingjustafewSNPs.Nevertheless,inasecondstepwetrytomodifytheindicatorfunctionto
limitthemultipletestingissueandimprovepower.Inaddition,weapplyourtestto10consensus
Alzheimer’sdisease(AD)susceptibilitySNPsthatwerepreviouslyidentifiedbyGWAS.
Categories: Association:Genome‐wide,Gene‐GeneInteraction
P47
Fine‐mappingeGFRsusceptibilitylocithroughtrans‐ethnicmeta‐
analysis
AnubhaMahajan1,JeffreyHaessler2,NoraFranceschini3,AndrewMorris4
1WellcomeTrustCentreforHumanGenetics,UniversityofOxford,Oxford,UK
2PublicHealthSciencesDivision,FredHutchinsonCancerResearchCenter,Seattle,Washington,USA
3UniversityofNorthCarolina,ChapelHill,NC,USA
4WellcomeTrustCentreforHumanGenetics,UniversityofOxford,Oxford,UK;DepartmentofBiostatistics,
UniversityofLiverpool,Liverpool,UK;EstonianGenomeCenter,UniversityofTartu,Tartu,Estonia
Reducedestimatedglomerularfiltrationrate(eGFR),isusedtodefinechronickidneydisease(CKD).
Weperformedtrans‐ethnicmeta‐analysistofine‐mapknowneGFRlocibyleveragingdifferencesin
distributionoflinkagedisequilibriumbetweendiversepopulations.Weconsideredsixgenome‐
wideassociationstudies(GWAS)comprisingof23,568individualsofEuropean,AfricanAmerican,
andHispanicancestry,eachsupplementedbyimputationuptothe1000GenomesProjectreference
panel(March2012release).Withineachstudy,associationwitheGFR(MDRDequation)wastested
underanadditivemodel.Wethencombinedassociationsummarystatisticsacrossstudieswith
MANTRA,500kbupanddownoftheleadSNPatknowneGFRloci,andconstructed“crediblesets”of
SNPsthatencompass99%oftheposteriorprobabilityofbeingcausal.Weresolvedfine‐mappingof
potentialcausalvariantstolessthan20variantsatthreeloci:GCKR(3SNPs,144.5kb),
UMOD/PDILT(4SNPs,39.3kb),andSHROOM3(19SNPs,74kb).AtGCKR,thecrediblesetcovers
threeSNPsincludingGCKRP446L,whichispredictedtobethefunctionalvariantatthislocus.
Variantsinthe99%crediblesetforSHROOM3,includeintronicvariantsinthegeneandoverlap
regulatoryelementsfromENCODE,therebyhighlightingapotentialmechanismfortheactionofthis
locusoneGFR.Thesefindingsprovideevidencethattrans‐ethnicGWAScanbeusedtofine‐map
potentiallycausalvariantsatcomplextraitslocithatcanbetakenforwardforexperimental
validationandcouldhelptofurtherourunderstandingofthebiologicalmechanismsunderlying
disease.
Categories: Association:Genome‐wide,FineMapping
P48
Arelipidriskallelesidentifiedingenome‐wideassociationstudiesready
fortranslationtoclinicalstudies?
Alexander M Kulminski1, Irina Culminskaya1, Konstantin G Arbeev1, Liubov S Arbeeva1, Svetlana V Ukraintseva1, Eric Stallard1, Anatoli I Yashin1 1
Duke University Insightsintogeneticoriginofdiseasesandrelatedtraitscouldsubstantiallyimpactstrategiesfor
improvinghumanhealth.Theresultsofgenome‐wideassociationstudies(GWAS)areoften
positionedasdiscoveriesofunconditionalriskallelesofcomplexhealthtraits.Were‐analyzedthe
associationsofSNPsdiscoveredascorrelatesoftotalcholesterol(TC)inalarge‐scaleGWASmeta‐
analysis.Wefocusedonthreegenerationsof9,167participantsoftheFraminghamHeartStudy
(FHS)whichwasapartofthatmeta‐analysis.WeshowedthatnoneofSNPsavailableintheFHShas
unconditionalriskallelesforTC.Instead,theeffectsoftheseSNPswereclusteredindifferentFHS
generationsinsex‐specificorsex‐unspecificfashion.Sensitivityoftheeffectstogenerationsimplies
theroleoftheenvironmentand/ortheage‐relatedprocesses.Astrikingresultwaspredominant
clusteringofsignificantassociationswiththestrongesteffectsintheyoungest3rdGeneration
cohort.Thisclusteringwasnotexplainedbythesamplesizeorprocedure‐therapeuticissues.The
effectclusteringinspecificpopulationgroupsmaystronglyaffectsamplesizesneededtodetect
genome‐widesignificance.Asanexample,theeffectsizeforrs1800562inthe3rdGenerationcohort
requiredaslittleasabout13,000subjectstoachievegenomesignificancewhereasthatin
comparablesampleoftheFHSoriginalandoffspringcohortsrequiredmorethan106subjects.The
resultsonclusteringoftheeffectsoflipidriskallelesareinlinewithexperimentalevidenceat
phenotypiclevelsfrompriorstudies.OurresultssuggestthatstandardGWASstrategiesneedtobe
greatlyexpandedtoefficientlytranslategeneticdiscoveriesintoclinicalstudies.
Categories: Association:Genome‐wide
P49
Genome‐widemeta‐analysisofsmoking‐dependentgeneticeffectson
obesitytraits:theGIANT(GeneticInvestigationofANthropometric
Traits)Consortium
AnneE Justice1, Thomas W Winkler2, Kristin L Young1, Jacek Czajkowski3, Nancy Heard‐Costa4,5, Mariaelisa Graff1, Xuan Deng6, Virginia Fisher6, Tuomas Kilpeläinen7, L Adrienne Cupples4,6 1
University of North Carolina at Chapel Hill, Department of Epidemiology, Chapel Hill, NC, USA Department of Genetic Epidemiology, Institute of Epidemiology and Preventive Medicine, University of Regensburg, Regensburg, Germany 3
Department of Genetics Division of Statistical Genomics, Washington University School of Medicine, St. Louis, MO, USA 4
NHLBI Framingham Heart Study, Framingham, MA , USA 5
Boston University, School of Medicine, Boston, MA, USA 6
Department of Biostatistics, Boston University School of Public Health, Boston University, Boston, MA, USA 7
The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark 2
Obesity and cigarette smoking (SMK) are important risk factors for cardiovascular disease. Yet, smokers often exhibit lower body mass index (BMI) and higher waist circumference (WC), and smoking cessation leads to weight gain. Genome‐wide association (GWA) studies have identified loci that are associated with risk of overall and central obesity; yet little is known about how SMK influences genetic susceptibility to obesity. This study aims to identify loci associated with obesity measured by BMI, WC adjusted for BMI (WCa), and waist to hip ratio adjusted for BMI (WHRa), and the influence of SMK on those genetic associations. We analyzed study specific association results from 88 studies including up to 210,153 subjects with GWA or Metabochip data. Each study employed two association models: 1) SNP effects adjusted for SMK (βadj), 2) SNP effects stratified by SMK. Study specific results were combined by inverse‐variance weighted fixed‐effects meta‐analyses. To detect SMK‐dependent genetic effects on obesity, the SMK‐stratified meta‐analysis results were used to calculate (i) the difference in SNP associations between current and non‐smokers (βdiff), and (ii) the joint estimates (βj) of the main effect and βdiff. We found genome‐wide significant (GWS) (p<5E‐8) evidence for non‐zero βdiff for two loci associated with WCa, three with WHRa and two with BMI. A total of 81 loci for WCa (14 are novel), 68 loci for BMI (10 are novel) and 50 loci for WHRa (nine are novel) reached GWS for βj and/or βadj. Our results highlight the importance of appropriately modeling genetic associations by considering known biological relationships between phenotypes and environment. Categories: Association: Genome‐wide, Cardiovascular Disease and Hypertension, Gene ‐ Environment Interaction
P 50 ABinomialRegressionModelforAssociationMappingofMultivariate
Phenotypes
Saurabh Ghosh 1, Arunabha Majumdar1 1
INDIAN STATISTICAL INSTITUTE, KOLKATA, INDIA Most clinical end‐point traits are governed by a set of quantitative and qualitative precursors and hence, it may be a prudent strategy to analyze a multivariate phenotype vector comprising these precursor variables for association mapping of the end‐point trait. The major statistical challenge in the analyses of multivariate phenotypes lies in the modelling of the vector of phenotypes, particularly in the presence of both quantitative and binary precursors. Likelihood based approaches such as variance components as well as data reduction techniques such as principal components become infeasible or biologically difficult to interpret if some of the components of the phenotype vector are qualitative in nature. We propose a Binomial regression approach that models the likelihood of the number of minor alleles at a SNP conditional on the vector of multivariate phenotype using a logistic link function. This framework allows for the integration of quantitative as well as binary phenotypes and does not require any distributional assumptions on the phenotype vector. The test for association is based on all the regression coefficients corresponding to the constituent phenotypes. The method can be easily adopted for analyzing longitudinal data. We carry out extensive simulations under a wide spectrum of genetic models of a multivariate phenotype vector and show that the proposed test is more powerful compared to analyzing a reduced phenotype based on the first principal component of the constituent phenotypes as well as separate univariate analyses of the different phenotypes. We apply our method to analyze a multivariate phenotype comprising homocysteine levels, Vitamin B12 levels and folate levels in a study on coronary artery disease. Categories: Association: Genome‐wide, Association: Unrelated Cases‐Controls, Multivariate Phenotypes, Quantitative Trait Analysis
P 51 HowtoincludechromosomeXinyourgenome‐wideassociationstudy
Christina Loley1, Inke R König1,2, Jeanette Erdmann2,3, Andreas Ziegler1,2,4 1
Institut für Medizinische Biometrie und Statistik, Universität zu Lübeck, Universitätsklinikum Schleswig‐Holstein, Campus Lübeck, Lübeck, Germany 2
DZHK (German Centre for Cardiovascular Research), Lübeck, Germany 3
Institut für Integrative und Experimentelle Genomik, Universitätsklinikum Schleswig‐Holstein, Campus Lübeck, Lübeck, Germany 4
Zentrum für Klinische Studien, Universität zu Lübeck, Universitätsklinikum Schleswig‐Holstein, Campus Lübeck, Lübeck, Germany In current genome‐wide association studies (GWAS), the analysis is usually focused on autosomal variants only, and the sex chromosomes are often neglected. Recently, a number of technical hurdles have been described that add to a reluctance of including chromosome X in a GWAS, including complications in genotype calling, imputation, and selection of test statistics. To overcome this, we provide a "how to" guide for analyzing X chromosomal data within a standard GWAS. Following a general pipeline for GWAS, we highlight the steps in which the X chromosome requires specific attention, and we give tentative advice for each of these. Through this, we show that by selection of sensible algorithms and parameter settings, the inclusion of chromosome X in GWAS is manageable. Closing this gap is expected to further elucidate the genetic background of complex diseases, especially of those with sex‐
specific features. Categories: Association: Genome‐wide, Association: Unrelated Cases‐Controls
P 52 Exomechipmeta‐analysistoidentifyrarecodingvariantsassociated
withpulsepressure
James P Cook1, Evelin Mihailov2, Nicholas GD Masca3, Fotios Drenos4, Helen Warren5, Martin D Tobin1, Louise V Wain1, Patricia B Munroe5, ExomeBP Consortium 1
Department of Health Sciences, University of Leicester, Leicester, United Kingdom Estonian Genome Center, University of Tartu, Tartu, Estonia 3
Cardiovascular Biomedical Research Unit, University of Leicester, Leicester, United Kingdom 4
Centre for Cardiovascular Genetics, University College London, London, United Kingdom 5
William Harvey Research Institute, Barts and the London School of Medicine and Dentistry, Queen Mary University of London, London, United Kingdom 2
Pulse pressure (PP) is a measure of arterial stiffness (calculated as the difference between systolic and diastolic blood pressure (BP)) which is a strong risk factor for cardiovascular disease and stroke. Large European genome‐wide association studies have already identified multiple common variants associated with PP, however common variants do not explain all of the heritability of BP traits. It has been hypothesised that some of the remaining heritability is explained by rare variants. The exome chip was designed to act as an intermediate step between cost‐effective whole genome SNP arrays, which predominantly measure common variation, and exome re‐sequencing approaches, which measure rare coding variation. The array includes ~250,000 mainly low frequency exonic variants. The ExomeBP consortium has been formed to analyse the exome chip for four BP traits: SBP, DBP, PP and hypertension, and comprises ~83,000 individuals from 31 different studies.We report a large scale single variant meta‐analysis of PP, including >150,000 polymorphic SNPs with minor allele frequency <1%. Results demonstrate replication of known pulse pressure loci as well as identification of novel loci not previously associated with blood pressure. Gene‐based analyses are also being performed. I will describe the methodological challenges in undertaking single variant and gene‐based meta‐analyses of exome chip data, such as distinguishing between monomorphic and missing variants across studies, the effect of transforming the phenotype and the advantages of different gene based methods, and outline our plans to boost sample size to ~400,000 through collaboration with other consortia. Categories: Association: Genome‐wide, Cardiovascular Disease and Hypertension, Quantitative Trait Analysis
P 53 Genome‐widesearchforage‐andsex‐dependentgeneticeffectsfor
obesitytraits:MethodsandresultsfromtheGIANTConsortium
Thomas W Winkler1, Mariaelisa Graff2, Anne Justice2, Llilda Barata3, Mary Feitosa3, Iris M Heid1, Ingrid Borecki3, Kari E North2, Zoltán Kutalik4, Ruth JF Loos5 1
Department of Genetic Epidemiology, University of Regensburg, Regensburg, Germany Department of Epidemiology, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina 27514, USA 3
Department of Genetics, Washington University School of Medicine, St Louis, Missouri 63110, USA 4
Department of Medical Genetics, University of Lausanne, 1005 Lausanne, Switzerland 5
The Charles Bronfman Institute for Personalized Medicine, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA 2
Obesity differs between men and women and changes over time. Previous genome‐wide association meta‐analyses (GWAMAs) revealed sexually dimorphic loci for waist‐hip ratio (WHR), but little is known whether genetic effects on obesity traits change with age. We thus conducted GWAMAs stratified by age (cut‐off at 50 years) and by sex, involving 110 studies (N>310,000) of European ancestry. Each study tested up to 2.8M HapMap imputed SNPs for association with BMI and WHR in four strata (men≤50, women≤50, men>50, women>50). Using the stratum‐specific estimates, we tested for age‐specific effects (G x AGE), sex‐specific effects (G x SEX), and for age‐specific effects that differ between men and women (G x AGE x SEX). Each of the three interaction tests was conducted with and without a‐priori filtering for the overall association. For BMI, our analysis yielded 15 loci with significant age‐difference, of which 11 showed a stronger effect in the younger group. For WHR, our analysis yielded 44 sexually dimorphic loci, of which 11 showed opposite effects and 28 showed an effect in women only. We did not identify any 3‐way G x AGE x SEX effects. Analytical power computations showed that our strategy (i) was well‐powered for any kind of 2‐way interaction (G x AGE, G x SEX) and for the most extreme 3‐way interaction (involving opposite effects across the four strata), but (ii) lacks power to find the most plausible 3‐way interactions (effects that are only present or only lacking in one of the four strata). Our results underscore the importance of age‐ and sex‐stratified analyses to further investigate the genetic underpinning for obesity traits and demonstrate that more refined methods will be needed to establish most plausible 3‐way interaction effects. Categories: Association: Genome‐wide, Gene ‐ Environment Interaction, Heterogeneity, Homogeneity, Sample Size and Power
P 54 Meta‐analysisofgene‐setanalysesbasedongenomewideassociation
studies
Albert Rosenberger1, Heike Bickeböller1, Christopher I Amos2, Rayjean J Hung3, Paul Brennan4 1
Universitätsmedizin Göttingen, Germany Geisel School of Medicine, US 3
Lunenfeld‐Tanenbaum Research Institute, Canada 4
International Agency for Research on Cancer, Lyon, France 2
Gene‐set analysis (GSA) methods are used as complementing approaches to genome‐wide association studies (GWAS). The single marker association estimates of a predefined set of genes are either contrasted to those of all remaining genes or to a null non‐associated background. To pool p‐values of several GSAs, it is important to take into account the concordance in the observed patterns of single marker association estimates. We propose an enhanced version of Fisher’s inverse χ²‐method META‐
GSA, but weighting each study to account for imperfect correlation between patterns. We investigated the performance of META‐GSA by simulating 500 GWAS with 500 cases and 500 controls at 100 SNPs. Wilcoxon’s rank sum test was applied as GSA for each study. We could demonstrate that META‐GSA has greater power to discover truly associated genes sets compared to simply pooling the p‐values. Under the H0, i.e. if there is no difference in the true pattern between the gene set of interest and the set of remaining genes, the results of both approaches are found to be almost without correlation. Thus, we recommend not relying on p‐values alone when combining the results of independent GSAs. Applying META‐GSA to pool results of four case‐control GWAS of lung cancer risk (Central European Study and the Toronto/SLRI Study; German Lung Cancer Study and the MDACC Study) revealed the pathway GO0015291 (“transmembrane transporter activity”) as significantly enriched with associated genes (GSA‐
method: EASE, p=0.0315 corrected for multiple testing). Categories: Association: Genome‐wide, Cancer, Case‐Control Studies, Pathways
P 55 Meta‐analysisofcorrelatedtraitsusingsummarystatisticsfromGWAS
Xiaofeng Zhu1, Tao Feng1 1
Department of Epidemiology and Biostatistics, Case Western Reserve University Genome wide association study (GWAS) has identified many genetic variants underlying complex traits. Many detected genetic loci harbor variants that associate with multiple, even distinct traits.Most current analysis approaches focus on single traits, even though the final results from multiple traits are evaluated together. Such approaches miss the opportunity to systemically integrate the phenome‐wide data available for genetic association analysis. In this study, we propose a general approach that can integrate association evidence from summary statistics of multiple traits, either correlated, unrelated, continuous or binary traits, which may come from the same or different studies. We allow for trait heterogeneity effects. Population structure and cryptic relatedness can also be controlled. Our simulations suggest that the proposed method has improved statistical power over single trait analysis in the most of cases we studied. We also applied our method to a large genome‐wide association study and identified multiple variants which were missed by a single trait analysis. Our method also provides a way to study a pleotropic effect. Categories: Association: Genome‐wide, Multivariate Phenotypes
P 56 StudyingtheEthnicDifferencesintheGeneticsofType2Diabetesusing
thePopulationSpecificHumanPhenotypeNetworks
Jingya Qiu1, Christian Darabos1, Jason H Moore1 1
Dartmouth College GWAS led to the discovery of 200+ SNPs at 150+ loci associated with type 2 diabetes mellitus (T2DM). It was also observed that East Asians develop T2DM at a higher rate, younger age, and lower BMI than their European ancestry counterparts. The reason behind this occurrence remains elusive. We constructed human phenotype subnetworks (HPSNs) based on ethnicity‐specific data to quantitatively analyze and visualize the disparities in genetic variants between different ethnic groups. Our identification of interethnic differences in the genetic variants associated with T2DM suggests the possibility of different pathways involved in the pathogenesis of T2DM amongst different populations. With comprehensive searches through the NHGRI GWAS catalog literature, we manually curated over 2,500 ethnicity‐specific SNPs associated with T2DM and 48 other related traits. The GWAS catalog usually reports the data combined over the initial and replication samples, across the different ancestries. Analysis of all‐inclusive data can be misleading, as not all variants are transferable across diverse populations. The extraction of ethnicity data allowed us to construct population‐specific HPSNs. We identified 99 SNPs highly significant to T2DM, most initially discovered in Europeans and replicated in East Asians, suggesting shared biological pathways. Of the 99 SNPs, however, 21 were specific to East Asian populations but impossible to replicate in other cohorts. Furthermore, many SNPs showed significant differences in studies of comparable size. For example rs2237892 in locus KCNQ1, a critical gene in insulin‐secreting INS‐1 cells, proved to be highly significant in East Asian population (p‐
Value=2.5E‐40) but not in Europeans (p=7.2E‐04). Categories: Association: Genome‐wide, Bioinformatics, Data Mining, Diabetes, Gene ‐ Gene Interaction, Pathways, Population Genetics, Prediction Modelling
P 57 HierarchicalBayesianModelintegratingsequencingandimputation
uncertaintyusingMCMCmethodforrarevariantassociationdetection
Liang He1, Janne Pitkäniemi1,2, Mikko J Sillanpää3,4, Antti P Sarin5, Samuli Ripatti5,6 1
Department of Public Health, Hjelt Institute, University of Helsinki, Finland Cancer Registry, Institute for Statistical and Epidemiological Cancer Research, Helsinki, Finland 3
Department of Mathematical Sciences, University of Oulu, Oulu FIN‐90014, Finland 4
Department of Biology and Biocenter Oulu, University of Oulu, Oulu FIN‐90014, Finland 5
Institute for Molecular Medicine Finland FIMM, University of Helsinki, Finland 6
Wellcome Trust Sanger Institute, UK 2
Next generation sequencing has led to the studies of rare genetic variants, which are thought to explain the missing heritability for complex diseases. Most existing statistical methods for RV association detection do not account for the presence of sequencing errors and imputation uncertainty, which can largely affect the power and perturb the accuracy of association tests due to rare observations of minor alleles. Some proposed methods that assign different weights based on genotype quality leads to the reduction of observations, and thus statistical power. We develop a hierarchical Bayesian approach to powerfully estimate the association between rare variants and complex diseases and account for genotype uncertainty from both whole‐genome sequencing and imputation data using MCMC method. Our integrated framework, which combines the misclassification model with shrinkage‐based Bayesian variable selection, estimates the association and predicts the low‐quality genotype simultaneously by borrowing the strength from priors and the rest of high‐quality data, and allows for dealing with sequencing and imputation data simultaneously. Sequencing quality information or imputation uncertainty is incorporated into the integrated framework to achieve the optimal power. We test the proposed method on simulated data and demonstrate that it outperforms other existing methods under various scenarios. Then we apply our model to a Finnish low‐density lipid cholesterol study, which includes both whole‐genome deep sequencing and imputation genotypic data, and both well‐known and novel gene regions with RVs significantly related to low density lipoprotein cholesterol level are identified. Categories: Association: Genome‐wide, Bayesian Analysis, Genomic Variation, Markov Chain Monte Carlo Methods, Missing Data, Multilocus Analysis, Population Genetics, Quantitative Trait Analysis, Sequencing Data
P 58 Sex‐specificassociationofMYLIPwithmortality‐optimizedhealthyaging
index
Mary F Feitosa1, Ryan L. Minster2, Mary K Wojczynski1, Jason L Sanders3, Amy M Matteini4, Richard Mayeux5, Nicole Schupf6, Thomas T Perls7, Kaare Christensen8, Anne B Newman3 1
Division of Statistical Genomics, Department of Genetics, Washington University School of Medicine, St. Louis, MO Department of Human Genetics, University of Pittsburgh, PA 3
Department of Epidemiology Graduate School of Public Health, University of Pittsburgh, PA 4
Division of Geriatric Medicine and Gerontology, School of Medicine, Johns Hopkins University, Baltimore, MD 5
Department of Neurology, Columbia University, New York, NY 6
Taub Institute, College of Physicians and Surgeons, Columbia University, New York, NY 7
Section of Geriatrics, Department of Medicine, Boston University, Boston School of Medicine and Boston Medical Center, MA 8
The Danish Aging Research Center, Epidemiology, University of Southern Denmark, and Department of Clinical Genetics and Department of Clinical Biochemistry and Pharmacology, Odense University Hospital, Odense, Denmark 2
Elevated low‐density lipoprotein (LDL) cholesterol is associated with increased risk of coronary artery disease, cognitive decline and dementia. Although these diseases predict mortality, knowledge of the relationship between dyslipidemia and its genetic contributors to mortality is limited. A mortality‐
optimized healthy aging index (HAI‐M) demonstrated accuracy to predict mortality. We hypothesized that SNPs from the GLGC Consortium (N=30) associated with LDL contribute to HAI‐M variability in 3,534 subjects from the Long Life Family Study. To create HAI, systolic blood pressure, pulmonary vital capacity, creatinine, fasting glucose, and modified‐mini‐mental‐status‐examination‐score, were scored as 0 (healthiest tertile), 1 (middle tertile), or 2 (unhealthiest tertile, and clinical cutoffs for glucose), and the sum produced an index ranging from 0 (healthiest) to 10 (unhealthiest). The HAI‐M was generated by applying regression coefficients from Cox proportional hazards models for death from the Cardiovascular Health Study to each component of the HAI. MYLIP‐rs3757354 (p=0.0001, beta=‐0.16±0.04) and APOH‐
rs1801689 (p=0.03, beta=0.23±0.10) were associated with HAI‐M in a stepwise regression model. Accounting for family structure using a mixed model, MYLIP‐rs3757354 was significantly associated with HAI‐M (p=0.001, beta=‐0.14±0.04). There were sex‐specific effects. MYLIP‐rs3757354 was significantly associated with HAI‐M in men (p=0.0003, beta=‐0.24±0.06), but not in women (p=0.23, beta=‐
0.07±0.06). MYLIP (6p23‐p22) encodes the E3 ubiquitin ligase myosin regulatory light chain‐interacting protein and promotes degradation of the LDLR, a process that may be relevant to healthy aging. Categories: Association: Genome‐wide, Genomic Variation, Multiple Marker Disequilibrium Analysis
P 59 Geneticdeterminantsofliverfunctionandtheirrelationshiptocardio‐
metabolichealth
Niletthi De Silva1, Debbie Lawlor1, Thomas Gaunt1, Abigail Fraser1 1
University of Bristol Introduction: Genome‐wide association studies have identified several common variants robustly associated with liver function tests, primarily ALT, AST, ALP, GGT, Bilirubin and Albumin. These phenotypes have been used as markers of liver damage, and there is evidence from observational studies that these are related to future adverse cardiometabolic health. However, it is unclear to what extent these associations are causal or confounded (in particular by alcohol consumption and general greater adiposity). Aims: To examine the association of metabochip variants with ALP, ALT, AST, GGT, Bilirubin and Albumin to determine whether these replicate published genome‐wide association (GWAS) findings and to identify any new variants robustly associated with these traits. To use Mendelian randomization study to test whether ALT, AST, ALP, GGT, Bilirubin and Albumin (markers of liver damage) causally influence CHD, stroke, type 2 diabetes and related continuos outcomes ‐ fasting glucose, fasting insulin, LDL, HDL, triglycerides, total cholesterol, SBP and DBP. Methods: We carried out metabochip‐wide meta‐analyses of ALT, AST, ALP, GGT, Bilirubin and Albumin to identify any novel variants associated with these traits. We then tested multiple common variants robustly associated with ALT, AST, ALP GGT, Bilirubin and Albumin (3, 2, 11, 17, 5, 5 SNPs respectively) against incident and prevalent diabetes, CHD, stroke events, and the related continuous outcomes in 5437 individuals from four prospective cohorts under the UCLEB consortium. Results: We replicated several previously established loci robustly associated with ALT, AST, ALP, GGT, Bilirubin and Albumin. In addition we identified two novel loci associated with ALP and AST in the ABO and PNPLA3 locus respectively at p<5x10‐8 . We now aim to replicate these two novel loci in an independent data set from the discovery cohort. In multivarable analyses adjusted for several potential confounders (i.e: smoking status, social class, alcohol, BMI and waist circumference) we replicated several observational associations reported previously. Indviduals carrying greater number of ALT, AST, ALP, GGT, Bilirubin and Albumin raising allles had increased levels of ALT, AST, ALP, GGT, Bilirubin and Albumin (p<0.001). There was evidence from instrumental variables analyses that ALT, AST GGT and Albumin causally reduce the risk of stroke: OR per log10 increase in ALT, AST, GGT was 0.04 [95%CI: 0.01, 0.11), 0.00 [95%CI: 00, 0.03], 0.21 [95%CI:0.10, 0.44] respectively and OR per one mg/dl increase in albumin was 0.45 [95%CI:0.35,0.58]. Conclusion: Markers of liver damage in particualr ALT, AST GGT and Albumin may causally influence the risk of stroke. Categories: Association: Genome‐wide, Cardiovascular Disease and Hypertension, Causation, Mendelian Randomisation
P 60 Variableselectionmethodforcomplexgeneticeffectmodelsusing
RandomForests
Emily R Holzinger1, Silke Szymczak1, James Malley2, Joan E Bailey‐Wilson1 1
Computational and Statistical Genomics Branch, National Human Genome Research Institute, National Institutes of Health 2
Center for Information Technology, National Institutes of Health Standard analysis methods for genome wide association studies (GWAS) are not robust to complex disease models, such as interactions between variables with small main effects. These types of effects could, in part, contribute to the heritability of complex human traits. Machine learning methods that are capable of identifying interactions, such as Random Forests (RF), are an alternative analysis approach. One caveat to RF is that there is no clear way to distinguish between probable true hits and noise variables based on the importance metric calculated. To this end, we have developed a novel variable selection method for RF that has three components: 1. A permutation procedure to calculate the RF importance score. 2. Null variance estimation method to create more meaningful thresholds for variable selection. 3. Recurrency to address noise in the results due to randomness of the method. First, we simulated datasets with various genetic models, including different levels of main and interaction effects. Next, we assessed the Type I error and power of the RF method and compared it to regression based methods. We further tested the performance of the variable selection method using a biological GWAS dataset. Our simulated data findings indicate that optimizing the selection threshold can greatly reduce the number of false positives in the selected variables. However, the optimal threshold is highly dependent on the underlying simulated genetic model. The recurrency aspect of the method assists in selecting the appropriate threshold. Additionally, the power to identify main effects is comparable to linear regression analyses with the correct main effect terms explicitly modeled. In the biological dataset, our method identifies a similar set of SNPs as linear regression. Future directions will involve testing and comparing methods for modeling the selected variables in a more interpretable fashion. Categories: Association: Genome‐wide, Bioinformatics, Data Mining, Gene ‐ Gene Interaction, Machine Learning Tools
P 61 Identificationofsharedgeneticaetiologybetweenepidemiologically
linkeddisorderswithanapplicationtoobesityandosteoarthritis
Jennifer L Asimit1, Kalliope Panoutsopoulou1, Eleanor Wheeler1, Sonja Berndt2, the GIANT consortium, the arcOGEN consortium, Andrew P Morris3,4, Inés Barroso1, Eleftheria Zeggini1 1
Wellcome Trust Sanger Institute National Cancer Institute, US National Institutes of Health 3
Wellcome Trust Centre for Human Genetics, University of Oxford 4
Department of Biostatistics, University of Liverpool 2
A common approach to a genetic overlap analysis of two traits involves comparing p‐values from the genome‐wide association study (GWAS) of each trait. However, p‐values do not account for differences in power, whereas Bayes’ factors do, and may be approximated using summary statistics. We use simulation studies to compare the power of frequentist and Bayesian approaches to overlap analyses, and to decide on thresholds for comparison between the two methods. It is empirically illustrated in single‐disease associations that BFs have a decreasing proportion of false positives (PFP) as study size increases. For a log10(BF) threshold Lq of 1.69 (R=type II error cost/type I error cost=2, p0 = Pr(no association at SNP)=0.99), the PFP decreases from 7.38×10‐4 (N=2,000 each cases/controls) to 3.37×10‐4 (N=20,000), while for p‐values the PFP fluctuates near the p‐value threshold a regardless of study size. In a preliminary overlap analysis of obesity (GIANT consortium) with OA (arcOGEN consortium), the number of signals is similar at comparable threshold levels between BFs and p‐values, though not always overlapping. For Lq=0.91 (R=type II error cost/type I error cost=12, p0=Pr(no association at variant)=0.99), there are 18 identified shared variants, and the comparable a levels of 0.003 and 0.004 result in 15 and 28 hits, respectively. The most notable difference is that the Bayesian list contains rs13107325 (in SLC39A8/ZIP8), a variant previously associated with obesity‐related phenotypes such as BMI and blood pressure, and animal studies have shown that the zinc‐ZIP8‐MTF1 axis regulates OA pathogenesis. We are pursuing replication of this finding. Categories: Association: Genome‐wide
P 62 Investigationofgeneticriskfactorsofverylowbirthweightinfants
withintheGermanNeonatalNetwork
Michael Preuß 1,3, Andreas Ziegler1,2, Egbert Herting3, Wolfgang Göpel3 1
Institute of Medical Biometry and Statistics, University at Lübeck, University Hospital Schleswig‐Holstein ‐ Campus Lübeck, Germany; 2
Center for Clinical Trials, University at Lübeck, Lübeck, Germany 3
Department of Pediatrics, University at Lübeck, Lübeck, Germany Very Low Birth Weight (VLBW) infants have substantially increased mortality and morbidity rates, but the factors influencing long‐term development are not well understood. The German Neonatal Network (GNN) was founded in 2009 to identify genetic, clinical and social factors influencing etiology and long‐
term development of VLBW. Clinical information includes oxygen demand, administration of surfactant, catecholamine, steroid hormones and bronchopulmonary dysplasia (BPD), brain haemorrhage (IVH), sepsis and death among others. The cohort size is 20,000, and the recruitment includes more than one quarter of all German VLBW per year. DNA samples from more than 9000 VLBW as well as buccal swabs from mothers have been collected from a total of 54 participating German hospitals. Approximately 2600 VLBW from GNN were genotyped on the Axiom™ Genome‐Wide CEU 1 Array, and replication was performed in another 4400 GNN VLBW. Results of the initial genome‐wide association study revealed genome‐wide significance (p <5E‐08) for several traits. An interesting finding is for the use of surfactant during hospital stay with an association to LINGO2 (lead SNP rs4878404, initial p = 5E‐06, replication one‐
sided p = 2.3E‐03). These results demonstrate that GNN is a unique resource for genetic and pharmacogenetic studies in VLBW. Categories: Association: Genome‐wide, Association: Unrelated Cases‐Controls
P 63 Artificialintelligenceanalysisofepistasisinagenome‐wideassociation
studyofglaucoma
Jason H Moore1, Casey S Greene1, Doug Hill1 1
Dartmouth College The genetic basis of primary open‐angle glaucoma (POAG) is not yet understood but is likely the result of many interacting genetic variants that influence risk in the context of our local ecology. We introduce here the Exploratory Modeling for Extracting Relationships using Genetic and Evolutionary Navigation Techniques (EMERGENT) algorithm as an artificial intelligence approach to the genetic analysis of common human diseases. EMERGENT builds models of genetic variation from lists of mathematical functions using a form of genetic programming called computational evolution. A key feature of the system is the ability to utilize pre‐processed expert knowledge giving it the ability to explore model space much as a human would. We describe this system in detail and then apply it to the genetic analysis of POAG in the Glaucoma Gene Environment Initiative (GLAUGEN) study that included approximately 1272 cases and 1057 controls. A total of 657,366 single‐nucleotide polymorphisms (SNPs) from across the human genome were measured in these subjects. Analysis using the EMERGENT framework revealed a best model consisting of six SNPs that map to at least six different genes. Two of these genes have previously been associated with POAG in several studies. The others represent new hypotheses about the genetic basis of POAG. All of the SNPs are involved in non‐additive gene‐gene interactions. Further, the six genes are all directly or indirectly related through biological interactions to the vascular endothelial growth factor (VEGF) gene that is an actively investigated drug target for POAG. This study demonstrates the routine application of an artificial intelligence‐based system for the genetic analysis of complex human diseases. Categories: Association: Genome‐wide, Bioinformatics, Data Mining, Gene ‐ Gene Interaction, Machine Learning Tools
P 64 Mutationscausingcomplexdiseasemayundercertaincircumstancesbe
protectiveinanepidemiologicalsense
Sabine Siegert1, Andreas Wolf2, David N Cooper3, Michael Krawczak2, Michael Nothnagel1 1
Cologne Center for Genomics, University of Cologne, Cologne, Germany Institute of Medical Informatics and Statistics, Christian‐Albrechts University, Kiel, Germany 3
Institute of Medical Genetics, Cardiff University, Cardiff, United Kingdom 2
Guided by the practice of classical epidemiology, research into the genetic basis of complex disease usually takes for granted the dictum that causative mutations are invariably over‐represented among affected as compared to unaffected individuals. However, employing various models of population history and penetrance, we show that this supposition is not true and that a mutation involved in the etiology of a complex disease can under certain circumstances be depleted rather than enriched in the affected portion of the population. Such mutations are ‘protective’ in an epidemiological sense and would often tend to be erroneously excluded from further studies. Our apparently paradoxical finding is due to the possibility of a negative correlation between complementary causative mutations that may arise as a consequence of the specifics of the population genealogy. This phenomenon also has the potential to hamper efforts to identify rare causative mutations through whole‐genome sequencing. Categories: Association: Genome‐wide, Association: Unrelated Cases‐Controls
P 65 Genome‐wideAssociationStudyIdentifiesSNPrs17180299andMultiple
HaplotypesonCYP2B6,SPON1andGSG1LAssociatedwithPlasma
ConcentrationsoftheMethadoneR‐andS‐enantiomerinHeroin‐
dependentPatientsunderMethadoneMaintenanceTreatment
Hsin‐Chou Yang1,2,3, Shih‐Kai Chu1,2,4, Sheng‐Chang Wang5, Sheng‐Wen Liu5, Ing‐Kang Ho5, Hsiang‐Wei Kuo5, Yu‐Li Liu5,6 1
Institute of Statistical Science, Academia Sinica Bioinformatics Program, Taiwan International Graduate Program, Academia Sinica 3
School of Public Health, National Defense Medical Center 4
Institute of Biomedical Informatics, National Yang‐Ming University 5
Center for the Neuropsychiatric Research, National Health Research Institutes 6
Department of Psychiatry, National Taiwan University Hospital and National Taiwan University College of Medicine 2
Although methadone metabolic pathway has been partially revealed there is still no report regarding genome‐wide association studies to characterize genetic mechanisms of the plasma concentrations of methadone R‐ and S‐enantiomer. We conducted the first genome‐wide association study to identify genes associated with the plasma concentrations of methadone R‐ and S‐enantiomer and their metabolites in a methadone maintenance cohort. We made a series of rigorous examinations in data quality control to remove poor samples and SNPs. We carried out genome‐wide single‐locus and haplotype‐based association tests for four quantitative traits, the plasma concentrations of methadone R‐ and S‐enantiomer and their metabolites, of 344 heroin‐dependent patients who were treated with methadone maintenance treatment in the Han Chinese population of Taiwan. We identified a significant SNP rs17180299 (p = 2.24×10‐8) which can explain 9.541% of the variation of the plasma concentration of methadone R‐enantiomer. We also identified 17 haplotypes on SPON1, GSG1L, and CYP450 genes associated with the plasma concentration of methadone S‐enantiomer. They can explain about one‐
fourth of variation of the plasma concentration of S‐methadone as a whole, where two significant haplotypes on CYP2B6 already explained 10.72% of the variation. In conclusion, we identified important SNP and haplotypes which contribute to genetic variation of plasma concentration.The results shed light on the genetic mechanism concerned with the metabolism of methadone maintenance treatment in heroin‐dependent patients. Moreover, the results are also potentially applicable to prediction of methadone dose and methadone‐related death. Categories: Association: Genome‐wide, Haplotype Analysis, Quantitative Trait Analysis
P 66 Anonparametricregressionapproachtotheanalysisofgenomewide
associationstudies
Pianpool Kirdwichai1, M Fazil Baksh1 1
University of Reading Recently there has been a move towards development of regression inspired methods for analysis of genomewide association studies of complex diseases. This is because multiple testing methods, such as Bonferroni correction, tend to impose stringent significance thresholds and consequently, unless the study is very large, can reliably identify only those genomic regions with very strong association signals. However many complex diseases are suspected result from the cumulative action of many loci each having a small effect, there is a high probability the association signals in such studies will in fact be moderate and extremely strong signals will be very rare. Although methods with higher power than the Bonferroni correction have been proposed, these tend to produce more false positive findings. This challenging problem of methodology that is more efficient than existing approaches but with false positive findings comparable with Bonferroni is addressed in this talk. A novel method based on nonparametric regression, capable of reliably identifying candidate regions of disease‐gene association in GWAS is developed and evaluated. The method is model‐free and establishes significance thresholds that inherently account for the LD structure in the data through a tuning parameter and assigned weights. A theoretically supported, computationally efficient method for obtaining the optimal tuning parameter is proposed and evaluated. Results of extensive evaluations and comparisons with existing methods show that the proposed approach is not only powerful but also lead to substantial reduction in false positive findings. The method is illustrated using data from the Wellcome Trust Case Control Consortium study of Crohn's disease. Categories: Association: Genome‐wide, Association: Unrelated Cases‐Controls
P 67 Geneticinsightsintoprimarybiliarycirrhosis–aninternational
collaborativemeta‐analysisandreplicationstudy
Heather J Cordell1, George Mells2, Gideon M Hirschfield3, Canadian/US/Italian/UK PBC Consortia, Carl Anderson4, Mike Seldin5, Richard Sandford2, Katherine A Siminovitch6 1
Newcastle University University of Cambridge 3
University of Birmingham 4
Wellcome Trust Sanger Institute 5
UC Davis 6
University of Toronto 2
Previous genome‐wide association studies (GWAS) of primary biliary cirrhosis (PBC) have confirmed associations at the human leukocyte antigen (HLA)‐region and identified 27 non‐HLA susceptibility loci. We undertook genome‐wide imputation and meta‐analysis of discovery datasets from the North American, the Italian and the UK GWAS of PBC, with a combined, post‐QC sample size of 2745 cases and 9802 controls. Following meta‐analysis, index single nucleotide polymorphisms (SNPs) at selected loci with PGWMA<2×10‐5 were genotyped in a validation cohort consisting of 3716 cases and 4261 controls. To prioritise candidate variants and genes at confirmed risk loci, we used the ENCODE and the 1000Genomes datasets to identify SNPs within regulatory elements and non‐synonymous SNPs in LD with the index variant (r2>0.8). We identified seven previously unknown risk loci for PBC. Functional annotation of these loci revealed SNPs within regulatory elements that are predicted to affect expression of DGKQ (4p16), PAM (5q14) and IL21R (16p12), that are strongly‐correlated to the index variant. Other candidate genes include IL12B (5q31), which forms part of the IL‐12 signalling cascade, and CCL20 (2q36), which is involved in chemo‐attraction of lymphocytes and dendritic cells towards epithelia and is expressed by TH17 cells originating from Foxp3+ T cells. Pathway analysis identified several highly plausible gene sets associated with PBC, including the IL‐12 and JAK‐STAT signalling pathways, and implicated several other immune processes in the pathogenesis of PBC, including innate immune processes (e.g. IFN‐α,β signaling). Categories: Association: Genome‐wide
P 68 GenesAssociatedwithLungCancer,ChronicObstructivePulmonary
Disease,orBoth
Jun She1,2, Bo Deng1,3, Jie Na1, Julie M Cunningham1, Zhifu Sun1, Jason A Wampfler1, Tanya M Petterson1, Paul D Scanlon1, Shuo Zhang1,4, Christine Wendt5 1
Mayo Clinic, MN, U.S.A. Zhongshan Hospital, Fudan University, Shanghai, People's Republic of China 3
Institute of Surgery Research, Daping Hospital, Third Military Medical University, Chongqing, People's Republic of China 4
Tulane University, New Orleans, LA 5
University of Minnesota and Veterans Administration Medical Center, Minneapolis, MN, U.S.A. 2
Background Genetic contribution to lung cancer (LC) or chronic obstructive pulmonary disease (COPD) remains unclear; COPD is considered an important LC precursor independent of tobacco smoke exposure. Over 300 candidate genes have been associated with COPD and/or LC. We conducted a comprehensive validation study to tease apart these candidate genes using genome‐wide single nucleotide polymorphism based analysis (SNP‐GWA) in a Caucasian population. Methods We tested 4491 SNPs in 304 candidate genes after redundancy analysis of linkage disequilibrium. The SNP‐GWA data that tested the association of these genes with LC and/or COPD consisted of 2484 subjects including LC only (n=612), LC and COPD (573), COPD only (537), and controls (762). The biological roles were elucidated by transcript expression quantitative trait loci (eQTL), differential mRNA expression between tumor and normal lung, and pathway analyses, along with allele‐specific risks, assessed by odds ratio (OR) and 95% confidence interval (CI). Results We validated 11 SNPs of 8 candidate genes (G1‐G8): 4 for LC with COPD (G1‐G4), 4 for LC from COPD (G1,2,5,6), 2 for COPD only (G1,7) and 1 for LC only (G8). A SNP in G1 was inversely associated with COPD without LC (OR=0.47; 95% CI, 0.31‐0.72) or with LC (OR=0.40; 0.27‐0.60), supported by eQTL of SNP‐alleletypes with mRNA levels in germline tissues (P=0.02) and differential expression of G1 (P<10‐5). A SNP in G2 was inversely associated with LC that developed from COPD patients (OR=1.65; 1.17‐1.78), with significant difference of G2 transcript levels in tumor and normal lung tissues (P=0.01). Conclusion We found 2 genes to be strongly associated with the risk of COPD and/or LC, indicating potential targets to intervene COPD and LC. Categories: Association: Genome‐wide, Association: Unrelated Cases‐Controls, Cancer, Genomic Variation, Multifactorial Diseases, Pathways, Quantitative Trait Analysis
P 69 Ageneralapproachforcombiningdiverserarevariantassociationtests
providesimprovedpoweracrossawiderrangeofgeneticarchitecture
Nathan L Tintle1, Brian Greco2, Allison Hainline3, Jaron Arbet4, Kelsey Grinde5, Alejandra Benitez6 1
Dordt College University of Michigan 3
Vanderbilt University 4
Winona State University 5
St. Olaf College 6
Brown University 2
In the wake of the widespread availability of genome sequencing data made possible by way of next‐
generation technologies, a flood of gene‐based rare variant tests have been proposed. Most methods claim superior power against particular genetic architectures. However, an important practical issue remains for the applied researcher—namely, which test should be used for a particular association study which may consider multiple genes and/or multiple phenotypes. Recently, tests have been proposed which combine individual tests to minimize power loss while improving the robustness to a wide range of genetic architectures. In our analysis, we propose an expansion of these approaches, by providing a general method that works for combining an arbitrarily large number of any gene‐based rare variant test—a flexibility typically not available in other combined testing methods. We provide a theoretical framework for evaluating our combined test to provide direct insights into the relationship between test‐test correlation, test power and the combined test power relative to individual testing approaches and other combined testing approaches. We demonstrate that our flexible combined testing method can provide improved power and robustness against a wide range of genetic architectures. We further demonstrate the performance of our combined test on simulated genotypes, as well as on a dataset of real genotypes with simulated phenotypes. We support the increased use of flexible combined tests in practice to maximize robustness of rare‐variant testing strategies against a wide‐range of genetic architectures. Categories: Association: Genome‐wide, Association: Unrelated Cases‐Controls, Case‐Control Studies, Genomic Variation
P 70 AMethodologicalComparisonofEpistasisModelingofHighOrderGene‐
GeneInteractionswithApplicationtoGeneticProfilingofPAInfection
amongCysticFibrosisPatients
Wenjiang Fu1, Mengtian Shen1, Shunjie Guan1 1
Michigan State University Recent studies of epistasis have been focusing on high order gene‐gene interactions, including the classification and regression trees (CART)‐based methods, the Mann‐Whitney U‐statistic methods, Bayesian epistasis association mapping (BEAM), gene‐based gene‐gene interaction tests, and gene‐based Multifactor dimensionality reduction (MDR). These methods have been developed to identify gene‐gene interactions in GWA studies of complex diseases and have been demonstrated to identify potential high order interactions. However, comparison of these methods and their computational capacity has not been fully studied. In this paper, we will compare these methods and apply them to an exome sequencing study of cystic fibrosis (CF). Although CF is a recessive Mendelian disease with a mutation in the CFTR gene, the disease manifestation is complex with the potential dysfunction of a number of organs and high mortality rate in early ages. About 80% CF patients develop pseudomonas aeruginosa (PA) infection, which leads to failure of the lung, liver, pancreas, intestine or other organs, resulting in breathing difficulty, CF associated liver diseases, diabetes, male infertility, and other disorders, and ultimate death in early age. It has been recently reported that genes (eg. DCTN4) other than the CFTR may also be associated with the PA infection among CF patients. We apply a number of methods to identify high order gene‐gene interactions for genetic profiling of PA infection using exome sequencing data of a case‐control study. We compare these methods in terms of the power, the profiling robustness and accuracy. We conclude that PA infection among CF patients can be profiled using a small number of genes with high accuracy. Categories: Association: Genome‐wide, Association: Unrelated Cases‐Controls, Bioinformatics, Case‐
Control Studies, Gene ‐ Gene Interaction, Sequencing Data
P 71 eQTLandpathwayanalysisonexpressionprofilesofacattlecross
Markus O Scheinhardt1, Bodo Brand2, Daisy Zimmer3, Norbert Reinsch3, Manfred Schwerin1,4, Andreas Ziegler1,5 1
Institute of Medical Biometry and Statistics, University Lübeck, Germany Institute for Genome Biology, Leibniz Institute for Farm Animal Biology, Dummerstorf, Germany 3
Institute for Genetics and Biometry, Leibniz Institute for Farm Animal Biology, Dummerstorf, Germany 4
Institute for Farm Animal Research and Technology, University Rostock, Germany 5
Center for Clinical Trials Lübeck, University Lübeck, Germany 2
In farm animal science, mapping of expression quantitative trait loci (eQTL) becomes increasingly important for studying molecular mechanisms of complex traits, such as milk production or carcass traits in cattle. We investigated 145 female animals from an F2 resource population derived from a cross between Charolais (beef cattle) and German Holstein (dairy cattle) founder breeds. SNP genotyping of 37204 SNP was accomplished using Illumina BovineSNP50 Beadchip, and gene expression profiles of 10069 adrenal cortex transcripts were obtained from Affymetrix GeneChip®Bovine v1 Array. The expression values were decorrelated by means of a sire‐dam model at which we adjusted for relatedness, age and season year of slaughtering. Residuals were used to perform the eQTL analysis. An adaptive location test was applied to adjust for varying degrees of skewness and tail length of the gene expression distributions. A total of 1048 eQTLs were identified which were associated with the expression of 641 adrenal cortex transcripts. Ingenuity pathway analysis of transcripts differentially expressed among genotypes highlighted molecular and cellular functions related to carbohydrate and lipid metabolism to be affected by eQTLs within the F2 cross population. Categories: Association: Genome‐wide, Gene Expression Arrays, Gene Expression Patterns, Pathways, Quantitative Trait Analysis
P 72 Evidenceforpolygeniceffectsintwogenome‐wideassociationstudiesof
breastcancerusinggeneticallyenrichedcases
Olivia Leavy1, Luigi Palla1, Julian Peto1, Douglas Easton2, Frank Dudbridge1 1
London School of Hygiene and Tropical Medicine University of Cambridge 2
Over recent years genome‐wide association studies have proven to be successful in finding associations between genetic variants and phenotypes. However, much of the heritability remains to be explained for complex diseases. Polygenic scoring allows testing for substantive polygenic effects among the markers that are not individually significant in GWAS. This has been successfully applied to many complex diseases, but to date has not been demonstrated in breast cancer. We studied two datasets: the UK2 study and the British Breast Cancer Study (BBCS), both containing women who have at least two close relatives that have developed breast cancer. In the BBCS dataset most of the cases have bilateral breast cancer. The disease prevalence applicable to these studies therefore will be lower than the general prevalence for breast cancer. Methods given by Dudbridge (2013) can be used to estimate the genetic variance explained by the entire GWAS using information on training and replication datasets. The training and replication datasets were created by internally splitting each of the BBCS and UK2 datasets. Using different values of the prevalence for familial breast cancer, these being lower than the prevalence of breast cancer, we estimated the genetic variance explained to be between 11.5% and 47.3% for the BBCS and at least 35.5% for the UK2 study. Given the low heritability of breast cancer, these values are larger than typically seen in complex diseases and seem to reflect the stronger genetic effects present in familial cases. This is the first significant association of genome‐wide polygenic scores for breast cancer and confirms the value of using genetically enriched cases in GWAS. Categories: Association: Genome‐wide, Cancer
P 73 DoBoundariesMatterforTiledRegression?
Alexa JM Sorant1, Heejong Sung1, Tae‐Hwi Schwantes‐An1, Alexander F Wilson1 1
Computational and Statistical Genomics Branch, NHGRI, NIH Current methods of analyzing today's vast quantities of genetic data include regression‐based variable selection methods producing linear models incorporating the chosen predictors. One such method, Tiled Regression, begins by considering separately relatively small segments of the genome called tiles, using stepwise regression to choose a set of independent significant SNPs, if any, within each tile and then combining them for further selection at higher levels. A natural way to define tiles is to create boundaries around recombination hotspots, so that genetic variants likely to be highly correlated due to linkage disequilibrium are initially considered together. However, such grouping may not be critical to the ultimate selection of genetic components of a trait model. To study the effects of alternative boundary definitions, we used a simulated mini‐GWAS genome including 306,097 SNPs in 4000 unrelated individuals, with two kinds of phenotypes generated for each. For examination of type I error we generated 2000 non‐genetic traits based on a normal distribution. For examination of power, we generated 2000 traits from a simple additive model of genetic effects contributed by 7 independent SNPs with locus‐specific heritabilities ranging from .0005 to .0108. We analyzed each trait with TRAP (v. 1.3) using several different tile boundary schemes, including the usual hotspot‐based definition, combining sets of ten consecutive tiles into larger tiles, and a definition based on a fixed length in base pairs corresponding to the average size of the original tiles. With analyses of 400 replicates completed, we observed virtually no difference in either type I error or power resulting from the different tile boundary definitions. Categories: Association: Genome‐wide, Multilocus Analysis, Prediction Modelling, Quantitative Trait Analysis
P 74 METAINTER:meta‐analysistoolformultipleregressionmodels
Tatsiana Vaitsiakhovich1, Dmitriy Drichel2, Christine Herold2, Andre Lacour2, Tim Becker2 1
Institute for Medical Biometry, Informatics and Epidemiology, University of Bonn German Center for Neurodegenerative Diseases (DZNE), Bonn 2
The need to summarize the results of related Genome‐wide association studies (GWAS) has encouraged rapid development of new meta‐analytic methods and tools. Application of the fixed or random effects models to single‐marker association tests is a standard practice. More complex methods involving multiple parameters have been used seldom in view of the absence of a respective meta‐analysis pipeline. Meta‐analysis based on combining p‐values can be applied to any association test. However, in order to be powerful, meta‐analysis methods for high‐dimensional models should incorporate additional information such as study‐specific properties of parameter estimates, their effect directions, standard errors and covariance structure. In this context, a method for the synthesis of linear regression slopes (MSRS) has been recently proposed in the educational sciences. We elaborate this method for multiple logistic regression models and introduce a software tool METAINTER, which implements MSRS for an arbitrary number of model parameters as well as three further meta‐analysis methods. METAINTER provides meta‐analysis p‐values and common parameter estimates of multiple regression models, and can be used to test the homogeneity of studies results. The software can directly be applied to analyze the results of single‐SNP tests, global haplotype tests, tests for and under gene‐gene or gene‐
environment interaction. Via simulations for two‐SNP models we have shown that MSRS has correct type I error and its power comes very close to that of the joint analysis of the entire sample. We support the results by a real data analysis of six GWAS of type 2 Diabetes. Categories: Association: Genome‐wide, Case‐Control Studies, Data Integration, Gene ‐ Gene Interaction
P 75 SuccessfulreplicationofGWAShitsformultiplesclerosisin10,000
Germansusingtheexomearray
Theresa Holste1, Dorothea Buck2, Antonios Bayas3, Thomas Bettecken4, Andrew Chan5, Sabine Fleischer6, Andre Franke7, Ralf Gold5, Christiane Grätz8, Christoph Heesen6, Karl‐Heinz Jöckel9, Bernd C Kieseier10, Tania Kümpfel11, Wolfgang Lieb12, Markus M Nöthen13, Friedemann Paul14, Vilmos Posevitz15, Martin Stangel16, Konstantin Strauch17,18, Björn Tackenberg19, Florian T Bergh20, Hayrettin Tumani21, Melanie Waldenberger22,23, Frank Weber24, Brigitte Wildemann25, Uwe Zettl26, Frauke Zipp8, Bertram Müller‐
Myhsok24, Heinz Wiendl15, Bernhard Hemmer2, Andreas Ziegler1,27 on behalf of the German Competence Network for Multiple Sclerosis (KKNMS) 1
Institut für Medizinische Biometrie und Statistik, Universität zu Lübeck, Universitätsklinikum Schleswig‐Holstein, Campus Lübeck, Lübeck, Germany 2
Klinikum rechts der Isar, Department of Neurology, Technische Universität München, Munich, Germany 3
Department of Neurology, Klinikum Augsburg, Augsburg, Germany 4
Max Planck Institute of Psychiatry, Munich, Germany 5
Neuroimmunologisches Labor, St. Josef‐Hospital, Universitätsklinikum der Ruhr‐Universität Bochum, Bochum, Germany 6
Klinik und Poliklinik für Neurologie, Universitätsklinikum Hamburg‐Eppendorf, Hamburg, Germany 7
Institut für Klinische Molekularbiologie, Christian‐Albrechts‐Universität zu Kiel, Germany 8
Klinik und Poliklinik für Neurologie, Universitätsmedizin der Johannes Gutenberg‐Universität Mainz, Mainz, Germany 9
Institut für Medizinische Informatik, Biometrie und Epidemiologie, Universitätsklinikum Essen, Essen, Germany 10
Neurologische Klinik, Heinrich‐Heine Universität, Düsseldorf, Germany 11
Institut für Klinische Neuroimmunologie, Ludwig‐Maximilians‐Universität München, München, Germany 12
Institut für Epidemiologie and Biobank popgen, Christian‐Albrechts‐Universität zu Kiel, Germany 13
Institut für Humangenetik, Universitätsklinikum Bonn, Bonn, Germany 14
NeuroCure Clinical Research Center, Charité ‐ Universitätsmedizin Berlin, Germany 15
Klinik für Allgemeine Neurologie, Universiätsklinikum Münster, Münster, Germany 16
Klinik für Neurologie, Medizinische Hochschule Hannover, Hannover, Germany 17
Institute of Genetic Epidemiology, Helmholtz Zentrum München – German, Research Center for Environmental Health, Neuherberg, Germany 18
Institute of Medical Informatics, Biometry and Epidemiology, Chair of Genetic Epidemiology, Ludwig‐Maximilians‐
Universität, Munich, Germany 19
Klinik für Neurologie, Philipps‐Universität Marburg, Marburg, Germany 20
Klinik und Poliklinik für Neurologie, Universitätsklinikum Leipzig, Leipzig, Germany 21
Klinik und Poliklinik für Neurologie der Universität Ulm, Ulm, Germany 22
Research Unit of Molecular Epidemiology, Helmholtz Zentrum München ‐ German Research Center for Environmental Health, Neuherberg, Germany 23
Institute of Epidemiology II, Helmholtz Zentrum München – German Research Center for Environmental Health, Neuherberg, Germany 24
Max‐Planck‐Institut für Psychiatrie, München, Germany 25
Neurologische Klinik, Universität Heidelberg, Heidelberg, Germany 26
Klinik für Neurologie und Poliklinik, Universitätsklinikum Rostock, Universität Rostock, Rostock, Germany 27
Zentrum für Klinische Studien, Universität zu Lübeck, Lübeck, Germany Background: Several genome‐wide association studies (GWAS) were conducted in the past few years to identify genetic variants associated with multiple sclerosis (MS). The objective of this study was the replication of observed findings using the exome array. Methods: 4,476 German MS cases and 5,714 German controls were genotyped using Illumina’s HumanExome v1‐Chip. Genotype calling was performed with Illumina’s Genome StudioTM Genotyping Module, followed by zCall. Results: Replication was successful for 9 regions beside the HLA region that are listed in the Catalog of Published Genome‐
Wide Association Studies as associated with MS. Criteria for replication were SNPs with p<10‐5 that were either identical to reported SNPs or in linkage disequilibrium with r2 > 0.8 to reported SNPs or were located in the reported gene. Many SNPs in various HLA genes reached genome‐wide significance (p<5x10‐8). Collapsing methods for rare variants gave similar results. Overall, replication of reported findings was possible using the exome array. One association identified in this study was not reported before in any previous GWAS. Specifically, we found genome‐wide significance to the gene MMEL1 which was found to be associated with MS in a candidate gene study by Ban (2010 Genes Immun 11:660‐
4) and SNPs in the vicinity (145 kb) to MMEL1 were identified by Sawcer (2011 Nature 476:214‐9). Conclusion: In this study, findings of previous GWAS could be replicated in a large German consortium using the exome array. This is especially important because the German population shows only low levels of population substructure and is therefore well suited for the investigation of complex diseases. Categories: Association: Genome‐wide
P 76 SharedGeneticEffectsUnderlyingAgeatMenarche,AgeatNatural
MenopauseandBloodPressure
Erin K Wagner1, Jin Xia1, Yi‐Hsiang Hsu2, Chunyan He1 1
Indiana University Richard M. Fairbanks School of Public Health Harvard Medical School 2
Age at menarche (AM) and age at natural menopause (ANM) are both associated with the risk of cardiovascular disease and its risk factors including blood pressure (BP). BP is known to increase rapidly during puberty, and early menarche is associated with elevated BP in adolescent and adulthood. BP is also known to increase more steeply around age at menopause. Earlier menopause is associated with higher blood pressure, although it is still unclear whether menopause accelerates BP increase or increased BP leads to earlier menopause. The observed synchronization between reproductive aging and BP development raises questions about the possibility of common regulating mechanisms shared by these processes. Using data from genome‐wide association studies, we performed a bivariate meta‐
analysis of these traits to identify genes with pleiotropic effects for AM, ANM and BP. We identified 6 novel loci at or near ARNTL (11p15.2), FTO (16q12.2), DCAKD (17q21.31), ZNF652 (17q21.32), 14q32.2 and 20q13.32 (intergenic regions) were associated with AM and BP (P<5X10‐8). For the bivariate analysis for ANM and BP, we found multiple variants within 200kb region at the 6p21.33 locus were significantly associated with ANM and BP. This region harbors genes including PRRC2A, BAG6, DDAH2, VW7, and HSPA1B. Our results suggest shared genetic effects for AM, ANM and BP. The findings may help improve the understanding of the genetic architecture and molecular mechanisms underlying these traits. Categories: Association: Genome‐wide, Multivariate Phenotypes
P 77 IdentificationofcombinedCommon‐andRare‐Geneticvariances
associatedwithrenalfunctioninHanChinese
Guanjie Chen1, Zhenjian Zhang2, Adebowale Adeyemo1, Yanxun Zhou2, Ayo Doumatey1, Guozheng Liu2, Amy Bentley1, Daniel Shriner1, Congqing Jiang2, Charles N Rotimi1 1
Center for Research on Genomics and Global Health, National Human Genome Research Institute, NIH, Bethesda, Maryland, USA 2
Suizhou Central hospital, Suizhou, Hubei, China The public health burden of Chronic Kidney Disease (CKD) is increasing in developing countries including China with an overall prevalence of 10.8% defined as eGFR less than 60 mL/min per 1∙73 m² or presence of albuminuria; thus, about 120 million Chinese have CKD. Both genetic and non‐genetic factors including economic status, area of residence, age, hypertension, diabetes and history of CVD contribute to the development of CKD. Here, we investigate the contribution of rare and common exonic variants to susceptibility to CKD by analyzing exome array data in 991 Han Chinese genotyped with the Affymetrix Axiom Exome Genotyping Arrays. A total of 64,397 SNPs that passed QC filters with minor allele count ≥ 5 within 17,266 gene sets were carried forward for analysis; 6,649 gene sets had common variants only (8802 SNPs), 8802 gene sets had both common and rare variants, and 1815 gene sets had only rare variants. Common variants analysis was implemented in PLINK assumed additive genetic model. The common and rare gene sets analysis was implemented in the Simultaneous Analyses of Common and Rare Variants in complex traits (SCARVAsnp) statistical package. Analyses were adjusted for age, sex, BMI, and hypertension status. We identified significant associations (pvalue<2.57×10‐6) in DIDO1, MOG, and GAB2, and suggestive significant association (pvalue<2.57x10‐5) in DNAH5, LAMC3, and TRAP1 with observed the lowest p value of 3.6×10‐10. We replicated seven of the sixteen and six of the ten genes reported to be associated with renal disease respectively in European and Chinese ancestry studies. These findings promise to provide novel insight into the genetic basis of CKD in Chinese and perhaps other human populations Categories: Association: Genome‐wide, Multilocus Analysis
P 78 Pathwayandgene‐geneinteractionanalysisrevealsnewcandidategenes
formelanoma
Myriam Brossard1, Shenying Fang2, Amaury Vaysse1, Qingyi Wei3, Hamida Mohamdi1, Marie‐Françoise Avril4, Mark Lathrop5, Jeffrey E Lee2, Christopher I Amos6, Florence Demenais1 1
INSERM, UMR‐946, Paris, France; Université Paris Diderot, Paris, France MD Anderson Cancer Center, Houston, Texas, USA 3
Department of Medicine, Duke University School of Medicine, Durham, USA 4
Hôpital Cochin, Université Paris Descartes, Paris, France 5
Genome Quebec Innovation Centre, McGill University, Montreal, Canada 6
Geisel College of Medicine, Dartmouth College, New Hampshire, USA 2
GWAS have identified 17 loci associated with melanoma, but these loci account for a small part of melanoma risk. These GWAS used single‐SNP analysis which may be underpowered to detect SNPs with small effect and/or interacting with other SNPs. To identify new candidate genes for melanoma risk, we combined pathway analysis and tests of gene‐gene interactions within melanoma‐associated pathways. Pathway analysis was based on the gene‐set enrichment analysis (GSEA) approach, using the Gene Ontology (GO) database. GSEA was applied to single‐SNP statistics obtained from melanoma GWAS of the MELARISK study (3,976 subjects) and MDACC study (2,827 subjects). To identify GO categories enriched in association signals, the false discovery rate (FDR) was computed using 100,000 SNP permutations. We tested all SNP‐SNP interactions within the identified GOs using INTERSNP. One million Hapmap3‐imputed SNPs were assigned to 22,000 genes, which were assigned to 316 Level 4‐GO categories. We identified 5 GOs with FDR≤5% in the two studies: response to light stimulus, regulation of mitotic cell cycle, induction of programmed cell death, cytokine activity and oxidative phosphorylation. A total of 110 genes were driving the enrichment signals in these GOs. Nine of these genes were found to occur frequently with melanoma‐related terms through PubMed mining, of which 5 are new candidates for melanoma risk (TP63, MAPK1, IL6, IL15, NDUFA2). Gene‐gene interaction analysis within each of the 5 identified GOs showed evidence for interaction for 4 SNP pairs (P≤10‐4 in MELARISK and replication at 5% in MDACC). Two of these pairs, CMTM7‐TNFSF4 (combined P=3x10‐7) and TERF1‐AFAP1L2 (combined P=2x10‐6), are biologically relevant. Funding: INCa_5982, LNCC, FRM Categories: Association: Genome‐wide, Cancer, Gene ‐ Gene Interaction, Multilocus Analysis, Pathways
P 79 Leveragingevolutionarilyconserved,celltype‐specific,regulatoryregion
datatodetectnovelSNP‐TFPIassociations
Jessica Dennis1, Alejandra Medina‐Rivera2, Vinh Truong1, Lina Antounians2, Pierre Morange3, David Trégouët4, Michael Wilson2, France Gagnon1 1
Dalla Lana School of Public Health, University of Toronto, Canada Genetics & Genome Biology Program, SickKids Research Institute, Toronto, Canada 3
Faculty of Medicine, University of the Mediterranean, Marseille, France 4
Université Pierre et Marie Curie, Paris, France 2
Low plasma levels of tissue factor pathway inhibitor (TFPI), a key regulator of the extrinsic coagulation cascade, increase the risk of venous and arterial thrombosis. TFPI plasma levels are highly heritable, but the genetics underlying this heritability are poorly understood. Genetic variants in evolutionarily conserved, cell type‐specific gene regulatory regions are important to complex traits. Incorporating this information in genome‐wide association studies (GWAS) may increase power. We experimentally ascertained regulatory regions in human and rat aortic endothelial cells (EC; a primary source of TFPI) using ChIP‐seq for epigenetic histone modifications and transcription factors. We then conducted a GWAS of SNPs associated with TFPI in 253 individuals from 5 French‐Canadian families ascertained on venous thrombosis (VT), prioritizing SNPs in these regulatory regions via stratified false discovery rate (sFDR) control. We tested SNPs with sFDR <0.25 for replication in 1170 French VT patients and, in both study samples, tested the significance of our prioritization scheme by comparing the median t‐statistic of prioritized SNPs and SNPs selected from comparable random regions. None of the 39 SNPs associated with TFPI in the discovery sample replicated at an FDR <0.05. Although our prioritization scheme did not help identify TFPI‐associated SNPs, defining novel approaches sFDR approaches is of great interest. Since TFPI is up‐regulated in inflamed vascular EC, we will next prioritize SNPs in experimentally determined inflammation‐specific vascular EC genes and their regulatory regions. Categories: Association: Genome‐wide, Bioinformatics, Data Integration, Epigenetic Data, Epigenetics, Sequencing Data
P 80 Asoftwarepackageforgenome‐wideassociationstudieswithRandom
SurvivalForests
Marvin N Wright1, Andreas Ziegler1,2 1
Institut für Medizinische Biometrie und Statistik, Universität zu Lübeck, Universitätsklinikum Schleswig‐Holstein, Campus Lübeck, Germany 2
Zentrum für Klinische Studien, Universität zu Lübeck, Universitätsklinikum Schleswig‐Holstein, Campus Lübeck, Germany In recent years, Random Forests have been successfully used to analyze genome‐wide association studies (GWAS) with dichotomous and quantitative endpoints. For censored survival endpoints software is available, namely Random Survival Forests and Conditional Inference Forests. However, due to computational burdens and memory issues, these tools are not capable of handling high‐dimensional data on GWAS scale. Consequently, we are not aware of any study applying one of them to genome‐wide data. We therefore introduce the new software package Random Jungle 3, which embeds the functionality of Random Survival Forests into the computationally efficient framework of Random Jungle. Compared to the original implementation, the runtime is reduced considerably, making the analysis of GWAS data possible. We validate the new software in extensive simulation studies. Finally, we apply it to a real dataset to assess the importance of involved single nucleotide polymorphisms (SNPs). Categories: Association: Genome‐wide, Bioinformatics, Data Mining, Machine Learning Tools
P 81 Identificationofnovelcommonandraregeneticvariantsassociatedwith
renalfunctioninHanChinese
Guanjie CHEN1, Zhenjian Zhang2, Adebowale Adeyemo1, Yanxun Zhou2, Ayo Doumatey1, Jie ZHOU1, Amy Bentley1, Daniel Shriner1, Charles Rotimi1 1
Center for Research on Genomics and Global Health, National Human Genome Research Institute, NIH, Bethesda, Maryland, USA 2
Suizhou Central hospital, Suizhou, Hubei, China The public health burden of CKD is increasing in developing countries including China with an overall prevalence of 10.8% defined as eGFR less than 60 mL/min per 1∙73 m² or presence of albuminuria; thus, about 120 million Chinese have CKD. Both genetic and non‐genetic factors including economic status, area of residence, age, hypertension, diabetes and history of CVD contribute to the development of CKD. Here, we investigate the contribution of rare and common exonic variants to susceptibility to CKD by analyzing exome array data in 991 Han Chinese genotyped with the Affymetrix Axiom Exome Genotyping Arrays. A total of 64,397 SNPs that passed QC filters with minor allele count ≥ 5 within 17,266 gene sets were carried forward for analysis; 6,649 gene sets had common variants only, 8802 gene sets had both common and rare variants, and 1815 gene sets had only rare variants. Common variants analysis was implemented in PLINK assumed additive genetic model. The common and rare gene sets analysis was implemented in the Simultaneous Analyses of Common and Rare Variants in complex traits (SCARVAsnp) statistical package. Analyses were adjusted for age, sex, BMI, and hypertension status. We identified significant associations (pvalue<2.57×10‐6) in DIDO1, MOG, and GAB2, and suggestive significant association (pvalue<2.57x10‐5) in DNAH5, LAMC3, and TRAP1 with observed lowest p value of 3.6 ×10‐
10. We replicated seven of the sixteen and six of the ten genes reported to be associated with renal disease respectively in European and Chinese ancestry studies. These findings promise to provide novel insight into the genetic basis of CKD in Chinese and perhaps other human populations. Categories: Association: Genome‐wide, Multilocus Analysis
P 82 AGenome‐WideAssociationStudytoExploreGene‐environment
InteractionwithParentalSmokingandtheRiskofChildhoodAcute
LymphocyticLeukemia
Jessica L Barrington‐Trimis1 1
University of Southern California, Keck School of Medicine, Department of Preventive Medicine Genetic susceptibility to parental smoking around pregnancy and risk of childhood acute lymphocytic leukemia (ALL) has not been fully explored. In this analysis, we used novel methods to scan the genome for gene‐parental smoking interactions. Participants were Hispanic cases and controls participating in the California Childhood Leukemia Study. Cases (N=380) were <15 years of age at diagnosis, and controls (N=454) were matched to cases on date of birth, gender, and maternal race. Genome‐wide genotyping was conducted using DNA from archival dried blood spot samples using the Illumina Human OmniExpress v.1 platform. Data were evaluated for the presence of multiplicative gene‐parental smoking interaction using statistically efficient two‐step scanning methods. We sought to replicate our most significant SNPs in two case‐only studies of childhood ALL in France (ESCALE, n=441), and Australia (AUS‐ALL, n=285). We identified two SNPs for replication for maternal smoking prior to and during pregnancy. One SNP was statistically significant in the AUS‐ALL replication, with the strongest results for maternal smoking during pregnancy, restricting to B‐cell progenitor ALL (summary interaction OR [CCLS/AUS‐ALL] = 4.40; 95% CI: 2.53, 7.64). Genotyping data for this SNP was not available in the ESCALE study. A second SNP was suggestive of a potential interaction in the AUS‐ALL replication (P=0.078, B‐cell ALL), but not in the ESCALE study where the interaction OR was in the opposite direction. Results indicate potential novel susceptibility loci for maternal smoking during pregnancy and risk of B‐cell ALL. Additional studies should be conducted to confirm these results in larger study populations of similar ethnic background. Categories: Association: Genome‐wide, Cancer, Case‐Control Studies, Gene ‐ Environment Interaction P 83 Network‐basedanalysisofGWASdata:Doesthegene‐wiseassociation
significancemodelingmatters?
Julie HAMON1, Yannick ALLANORE2, Maria MARTINEZ1 1
INSERM UMR1043, Hôpital Purpan, Toulouse INSERM 1016, Hôpital Cochin, Paris 2
Integrating prior biological knowledge into Genome‐Wide Association data may unravel sets of genes having collectively or in interaction a role on the disease. Several network‐based approaches have been proposed depending on the type of known information that is used to combine the genes such as protein‐protein interaction (PPI) network or gene functions pathways. These studies rely on the association of each gene with the disease, i.e., on an individual Gene‐Wise P‐value (GWP) which can be derived under different alternatives. Here, we aim to compare such different strategies in our Systemic Sclerosis GWAS data. We built a two‐stage network study by randomly splitting our GWAS data into a scan and a replication dataset. In the scan dataset we performed a PPI network‐based approach using a dense module search strategy with different GWP values: for instance using the smallest single‐SNP P‐value either unadjusted (Min) or Bonferroni‐adjusted (Bonf) or using the Fisher’s method to combine all single‐SNP P values.The results were compared according to the length (number of genes) of the enriched sub‐modules and the characteristics of their genes. The top (5 and 10%) most enriched modules were tested for enrichment analysis in the replication dataset. We finally mapped the genes from the replicated sub‐networks to KEGG pathways. Overall, we found low consistency across the results from the different strategies: different sets of genes are selected but also different KEGG pathways are identified. Categories: Association: Genome‐wide, Gene ‐ Gene Interaction, Pathways
P 84 Heritabilityestimatesandgeneticassociationfor60+complextraitsina
younghealthysiblingcohort
Jun Z Li1, Qianyi Ma1, Ayse B Ozel1, Karl C Desch2, David Ginsburg3 1
Department of Human Genetics, University of Michigan, Ann Arbor Department of Pediatrics and Communicable Disease, University of Michigan, Ann Arbor 3
Howard Hughes Medical Institute, Department of Internal Medicine, University of Michigan, Ann Arbor 2
As genotyping becomes more efficient, sample recruitment and phenotyping remain a major limiting factor. In a GWAS of bleeding and blood clotting traits we sought to increase the utility of the cohort by collecting > 60 self‐reported complex traits through web‐based questionnaires. The cohort of 1,191 healthy young subjects consists of 509 sibships, 80% Europeans, and age of 14‐35 yrs. The traits include 16 quantitative traits (e.g., weight, height, age of menarche, hematological measures RBC, HCT, MCV, MCH, MCHC, RDW, WBC, HGB, PLT, MPV), 21 ordinal traits (e.g., Smoking, BleedingTendency, SkinTags, Acne, TanningTendency, SkinColor, Freckles, DentalCaries, VisionCorrection, EatingSweets, EatingSaltyfood, Athleticability, Aphthousulcers), and 27 nominal traits (e.g., Immunization, ToothExtraction, EyeColor, HairColor, Hairline, EarLobeCreased, EarLobeAttachment, Dimples, Dyslexia, Migraines, Stuttering, Allergies, Flatfeet, Handedness, PhoticSneeze, BrainFreeze, InterlockingFingers, etc.). We used the known relatedness to estimate heritability using Merlin‐regress and found that >1/2 of the traits have H^2 > 40%. Since the samples have been genotyped over ~800K SNPs in the original GWAS we used SNP data to calculate the actual genetic relatedness, and estimated the variance explained by all the genotyped SNPs using GCTA. With all subjects, pedigree‐based estimates were similar to SNP‐based estimates; but the latter were often reduced when we select one subject from each sibship to analyze the unrelated subsets. For many traits we identified common variants of significant association. This study demonstrates the feasibility of simultaneous analysis of dozens of traits via web‐
based profiling. Categories: Association: Genome‐wide, Heritability, Multivariate Phenotypes, Quantitative Trait Analysis
P 85 Large‐scaleexomechipgenotypingrevealsnovelcodingvariation
associatedwithendometriosis
Andrew P Morris1, Reedik Mägi2, Nilufer Rahmioglu3, Anubha Mahajan3, Neil Robertson3, Marie Peters4, Merli Saare4, Andres Salumets4, Krina T Zondervan3 1
Department of Biostatistics, University of Liverpool, Liverpool, UK Estonian Genome Centre, University of Tartu, Tartu, Estonia 3
Wellcome Trust Centre for Human Genetics, University of Oxford, Oxford, UK 4
Department of Obstetrics and Gynaecology, University of Tartu, Tartu, Estonia 2
Genome‐wide association studies have identified nine loci harbouring common variants implicated in endometriosis, which together explain only ~3% of the heritability of the condition. To investigate the contribution of coding variation to endometriosis pathogenesis, we undertook genotyping with the Illumina Exome Chip of two studies of European ancestry: (i) 910 cases from the Oxford Endometriosis Gene study and 13,334 population controls from the UK Exome Chip Consortium; and (ii) 326 cases and 711 population controls from the Estonian Biobank. Within each study, we evaluated the association of endometriosis with: (i) individual coding variants; and (ii) burden/over‐dispersion of loss of function (all frequencies) and rare non‐synonymous (minor allele frequency [MAF] less than 1%) variants within genes using SKAT‐O. Association summary statistics were combined across studies by meta‐analysis. We conducted pathway analysis on the basis of single variant meta‐analysis summary statistics using MAGENTA. No individual coding variants achieved exome‐wide significant evidence of association (p<5x10‐7, Bonferroni correction for 100,000 variants). The strongest signals include missense variants in TAF1L (D141N, p=1.5x10‐5, MAF=0.077%) and BMP3 (Y67N, p=3.2x10‐5, MAF=2.7%). We observed exome‐wide significant evidence of association (p<2.5x10‐6, Bonferroni correction for 20,000 genes) with burden/over‐dispersion of loss of function variants in C16orf89 (p=1.1x10‐6) and rare non‐
synonymous changes in NECAB3 (p=1.7x10‐7), ZNF485 (p=1.1x10‐6), and RSAD2 (p=2.1x10‐6). MAGENTA analyses highlighted potential involvement of cell adhesion/structure, immune function and cancer‐related pathways in endometriosis. Categories: Association: Genome‐wide, Case‐Control Studies
P 86 DissectingtheObesityDiseaseLandscape:IdentifyingGene‐Gene
InteractionsthatareHighlyAssociatedwithBodyMassIndex(BMI)
Rishika De1, Shefali Setia Verma2, Sarah Pendergrass2, Fotios Drenos3, Michael Holmes4, Folkert Asselbergs5, Brendan Keating4, Marylyn Ritchie2, Diane Gilbert‐Diamond1 1
Dartmouth College Pennsylvania State University 3
University College London 4
University of Pennsylvania 5
University Medical Center Utrecht 2
Though obesity is estimated to have a heritability of 40‐70%, less than 2% of its variation is explained by the BMI‐associated loci that have been identified so far. Hence, interactions between genes, i.e. epistasis, may explain a larger portion of the heritability of BMI. We analyzed genetic information from 18,686 individuals across 5 cohorts – ARIC, CARDIA, FHS, CHS, MESA – to identify interactions between SNPs (Single Nucleotide Polymorphisms). Participants were genotyped using a targeted approach via the gene‐centric IBC array (ITMAT‐Broad‐CARe). SNPs were filtered using two parallel approaches – one based on the strength of their main effects of association, and the other a knowledge‐based approach called Biofilter that identifies biologically plausible SNP‐SNP models. Filtered SNPs were analyzed using QMDR (Quantitative Multifactor Dimensionality Reduction) to detect SNP‐SNP interactions that are highly associated with BMI. QMDR is a nonparametric, genetic model‐free method that detects non‐
linear interactions in the context of a quantitative trait. We identified 6 novel interactions with a Bonferroni corrected p‐value of association < 0.05. These interactions also replicated previously identified BMI‐associated independent signals ‐ rs12617233 in FLJ30838, rs997295 in MAP2K5, and rs1799998 in CYP11B2. Our results highlighted interactions between genes involved in mitochondrial dysfunction (POLG2), aldosterone synthase functioning (CYP11B2), cell proliferation (MAP2K5), insulin resistance (IGF1R, CAV3), vascular development (MAP2K5, EZR), cell adhesion (EZR) and apoptosis (EZR). This study highlights a novel approach to discovering gene‐gene interactions within the obesity disease landscape. Categories: Association: Genome‐wide, Bioinformatics, Gene ‐ Gene Interaction, Genomic Variation, Quantitative Trait Analysis P87
InvestigationofParent‐of‐OrigineffectsinAutismSpectrumDisorders
SiobhanConnolly1,ElizabethAHeron1
1TrinityCollegeDublin
Thedetectionofparent‐of‐origineffectsaimstoidentifywhetherornotthefunctionalityofalleles,
andinturnassociatedphenotypictraits,dependsontheparentaloriginofthealleles.Genome‐Wide
AssociationStudies(GWAS)havehadlimitedsuccessinexplainingtheheritabilityofmanycomplex
disordersandtraitsbutsuccessfulidentificationofparent‐of‐origineffectsusingtrio(mother,
father,offspring)GWASmayhelpshedlightonthismissingheritability.AutismSpectrumDisorders
(ASDs)areconsideredtobeheritableneurodevelopmentaldisordersandanumberoftrioGWAS
datasetsexistforexaminingthisheritability.Here,wehaveinvestigatedparent‐of‐origineffectsin
largetrioGWASdatasetsthathavepreviouslybeenanalysedforparent‐of‐origineffectsusing
statisticalapproachesthatdidnothavethecapacitytodetectepigeneticeffectssuchasmaternal‐
offspringgeneticeffectsandallassumptionsoftheapproachesmaynothavebeensatisfied.Here
theapproachofEstimationofMaternal,ImprintingandInteractionEffectsUsingMultinomial
Modelling(EMIM)isusedtoidentifySNPsassociatedwithASDthroughaparent‐of‐origin
mechanismwhichhasthepotentialtoaidinunderstandingmorefullythegeneticunderpinningsof
ASD.
Categories: Association:Genome‐wide,PsychiatricDiseases,TransmissionandImprinting
P88
IntegrativeclusteringofmultiplegenomicdatausingNon‐negative
MatrixFactorization
PrabhakarChalise1,BrookeLFridley1
1UniversityofKansasMedicalCenter,KansasCity,KSUSA
Weproposeanovelapproachforintegrativeclusteringofmultiplegenomicdatasetstoclassifythe
diseasesubtypes.ThemethodusesNon‐negativeMatrixFactorization(NMF)techniqueby
extendingtheexistingmethodforsingledatainordertoutilizethestrengthsacrossmultipledata
types.Thisanalysisapproachwasappliedtothecancergenomeatlas(TCGA)studiesonovarian
cancerinvolving499subjectsthathavebothgeneexpression(90797probes)andmethylation
(27338probes)assaysontumorsamplesavailable.Togetclinicallymeaningfulclusters,top500
mostassociatedprobesfromeachdatasetwereselectedbyfittingcoxproportionalhazardsmodel
withtimetorecurrence(TTR)ofthediseaseasendpointforeachprobeadjustingforageand
cancerstage.Theintegrativemethodresultedinthreeoptimumclustersofsamples.Thephenotypic
differencesofTTRamongtheseclusterswereassessedbyKaplanMeierplotfollowedbylogrank
test(p=1.0×10‐11).Further,eachexpressionandCpGprobewasassessedacrossthethreeclusters
usinganalysisofvariancefollowedbymultipletestingadjustments(BenzaminiandHochberg).
Amongothersignificantprobes,thegenesTXNDC9(p=7.23×10‐35)andCRBN(p=8.44×10‐32)and
theCpGprobesneargenesPLOD2(p=1.91×10‐4)andKRTAP11‐1(p=3.24×10‐4)werefoundtobe
mostsignificantlydifferentacrosstheclusters.Furtherstudiesareneededtodeterminethe
functionalrelevanceofthesegenesintheovariancanceretiology.
Categories: Association:Genome‐wide,Cancer,DataIntegration
P89
Toolsforrobustanalysisingenome‐wideassociationstudiesusing
STATA
NikiDimou1,PantelisBagos1
1UniversityofThessaly
Withinthecontextofgeneticassociationstudies(GAS)andgenome‐wideassociationstudies
(GWAS)thereisavarietyofstatisticaltechniquesinordertoconducttheanalysisbutacommon
problemisthelackofknowledgeconcerningthemodelofinheritance.Severalapproacheshave
beenproposedforderivingrobustproceduresthatwilldetectthetrueunderlyingmodelof
inheritanceand,atthesametimeperformtheanalysismaximizingthepowerandpreservingthe
nominaltypeIerrorrate.Theprimarygoalofthisworkistoimplementasmanyaspossiblerobust
methodswithinthestatisticalpackageSTATAandsubsequentlytomakethesoftwareavailableto
thescientificcommunity.RobustmethodsbasedontheMAXstatistic,theMERTstatistic,theMIN2,
aswellastheGMSandtheGMEprocedureswereimplementedinSTATAandimmediatecommands
wereconstructed.Themaindifficultyinimplementingtheabove‐mentionedmethodsisthefactthat
theyarecomputationallyintensivesince(withtheexceptionofMERT)theasymptoticpropertiesof
theestimatorscannotbederivedanalyticallyandothermethodsareneeded.ConcerningMAX,GMS
andGME,weusedaseveralfastMonteCarlosimulationmethodsinordertocalculateaccuratep‐
values,whereasforMIN2,wereliedonnumericalintegration.Thisisthefirstcompleteeffortto
implementproceduresforrobustanalysisandselectionoftheappropriategeneticmodelinGASor
GWASusingSTATA.Sincethereareonlyafewavailablesoftwareimplementationsoftherobust
methodsformeta‐analysisofGASorGWASourfuturegoalistoextendoursoftwareinthecontext
ofmeta‐analysisusingSTATA.Thesoftwareisavailableathttp://www.compgen.org/tools/robust‐
meta‐analysis.
P90
Developmentofathree‐waymixedmodellingapproachintegrating
geneticandclinicalvariablesinanalysisofearlytreatmentoutcomesin
epilepsy.
BenFrancis1,AndreaJorgensen1,AndrewMorris1,AnthonyMarson1,MichaelJohnson2,Graeme
Sills1,EpiPGXconsortium
1UniversityofLiverpool
2ImperialCollegeLondon
Remissionfromseizures(12monthsofseizurefreedom)isindicativeoftherapeuticresponsetoan
antiepilepticdrug(AED)whentreatingepilepsypatients.Clinicalfactorsincludinggenderand
epilepsytypehavebeenattributedtotheremissionoutcomeandpotentialpharmacogeneticfactors
arenowbeinginvestigated.
Atotalof964patientsfromtheStandardandNewAntiepilepticDrug(SANAD)study,arandomised
trialthatcomparedtreatmentswithvariousAEDsinpatientswithnewlydiagnosedepilepsy,were
genotypedtoinvestigategeneticbiomarkersfortimetoremission,aswellasotherlongitudinal
phenotypes,includingtimetoAEDwithdrawalandtimetofirstseizure.Analysiswasinitially
undertakenusingatraditionalonecomponentsurvivalmodelfortimetoremission,however,no
genome‐widesignificantSNPswerefound.
Thismethodmaylackpowerasthepopulationisconsideredhomogeneous.Thepresenceofthree
sub‐populationsfortimetoremissionisapparent;thosewhoexperienceremissionimmediately,
thosewhoexperienceremissioneventuallyandthosewhodonotexperienceremissionatanypoint
duringfollow‐up.Toconsiderthesesub‐populations,athreecomponentmodelisrequiredfor
survivalanalysis.Mixturemodellingwithacurefractionwasselectedastheoptimalmethodology
toderiveathreecomponentmodel.
Thefurtheradaptedmethodologyproposedinthisabstractwillbeappliedtoalargerpopulationof
patientswithnewly‐diagnosedepilepsythatisnowavailableviatheEpiPGXconsortium
(www.epipgx.eu),andwhichincludestheSANADcohortaswellasothercohortsofpatientsbeing
collectedworldwidetoinvestigategeneticbiomarkersofepilepsy.
Categories: Association:Genome‐wide,MultivariatePhenotypes,PopulationStratification,Prediction
Modelling,PsychiatricDiseases
P91
Meta‐analysisoflowfrequencyandrarecodingvariantsandpulmonary
function.
VictoriaEJackson1,LouiseVWain1,IanSayers2,IanPHall3,MartinDTobin1,SpiroMetaConsortium
1DepartmentofHealthSciences,UniversityofLeicester,Leicester,UnitedKingdom
2DivisionofRespiratoryMedicine,UniversityofNottingham,Nottingham,UnitedKingdom
3DivisionofTherapeuticsandMolecularMedicine,UniversityofNottingham,Nottingham,UnitedKingdom
Pulmonaryfunctionmeasuresareanimportantpredictorofmortalityandmorbidityandareused
inthediagnosisofanumberofdiseases,includingchronicobstructivepulmonarydisease(COPD).A
numberoflarge‐scalegenome‐wideassociationstudies(GWAS)havesuccessfullyidentifiedsingle
nucleotidepolymorphisms(SNPs)influencingpulmonaryfunctionin26regions;howevertheseso
faridentifiedregionsonlyaccountforasmallproportionoftheestimatedheritability.One
hypothesisisthattheso‐called“missingheritability”mightbefoundinrarevariantswithlarge
effects.Agenotypingarrayhasrecentlybeendevelopedasacost‐effectivewaytoinvestigatethe
effectsofrarevariantsinlargesamplesizes.Thevariantsincludedinthearraydesignwereselected
astheywereobservednumeroustimesinthesequencedexomesorgenomesofaset12,000
individualsfrom16samplecollectionsandarepredominantlylowfrequencyandrareexonicSNPs.
Wecarriedoutameta‐analysisofexomearraydataandthreepulmonaryfunctionmeasures(FEV1,
forcedvitalcapacity(FVC)andtheratioofFEV1toFVC(FEV1/FVC))inover30,000individualsof
Europeanancestry,from12studies,whohadbeengenotypedusingtheIlluminaHumanExome
beadchip.Wehaveutilisedsinglevariantassociationanalysismethods,traditionallyemployedin
GWAS,alongwithgene‐basedmethods,whichforthejointeffectofseveralvariantsinagene;the
lattermethodisconsideredamorepowerfulapproachtoidentifyrarevariantsassociatedwitha
trait.Wepresentemergingfindingsfromtheseanalyses.
Categories: Association:Genome‐wide,QuantitativeTraitAnalysis
P92
UsingPolygeneScoresandGCTAtoIdentifyaSubsetofSNPsthat
ContributetoGeneticRisk
ElizabethAHeron1,AlisonKMerikangas1,RicardoSegurado2
1DepartmentofPsychiatry&NeuropsychiatricGeneticsResearchGroup,TrinityCollegeDublin,Dublin2,
Ireland
2CentreforSupportandTraininginAnalysisandResearch,UniversityCollegeDublin,Dublin4,Ireland
Polygenescores1areameansofsummarisingthecombinedeffectofagroupofmarkers,inthis
casesinglenucleotidepolymorphisms(SNPs),thatasindividualmarkersperhapsdonotreach
statisticalsignificanceinagenome‐wideassociationstudy(GWAS),butinaggregateareassociated
withcasestatus.Thepolygenicscoringmethodoffersameansbywhichareducedsetofmarkers
canbeidentifiedthatoffergoodpredictionforaparticulartraitandcanperhapsnarrowthefocusof
theassociatedgeneticriskfactors.Genome‐wideComplexTraitAnalysis(GCTA)2isamethodby
whichtheproportionofphenotypicvariancethatisexplainedbySNPscanbeestimated.Thus,a
givensetofSNPscanbecomparedwithanothersetofSNPstodeterminewhichsetexplainsmore
ofthegeneticcomponentofthevariabilityinthephenotype.Theaimofthispaperistocombine
thesetwomethodologiestoidentifyasubsetofSNPsthatbothcontributesignificantlytothegenetic
componentofthephenotypicvariancebutthatalsooffergoodpredictionforaphenotypictraitof
interest.AnumberofGWASdatasetstogetherwithsimulateddatawillbeusedtoexplorethis
approachwhichoffersthepotentialtoaidinthedifficulttaskofidentifyingriskvariantsfor
complexdisorders.1.PurcellSM,WrayNR,StoneJL,VisscherPM,O’DonovanMC,SullivanPF,etal.
Commonpolygenicvariationcontributestoriskofschizophreniaandbipolardisorder.Nature
2009;460:748–52.2.YangJ,LeeSH,GoddardMEandVisscherPM.GCTA:atoolforGenome‐wide
ComplexTraitAnalysis.AmJHumGenet.2011Jan88(1):76‐82.
Causation
P93
ChallengingIssuesinGWASofHumanAgingandLongevity
AnatoliyIYashin1,DeqingWu1,KonstantinArbeev1,AlexanderKulminski1,LiubovSArbeeva1,
SvetlanaVUkraintseva1
1DukeUniversity
AnatoliyI.Yashin,DeqingWu,KonstantinG.Arbeev,LiubovS.Arbeeva,AlexanderM.Kulminski,
SvetlanaV.UkraintsevaDuringlastdecadesubstantialprogressingeneticanalysesofcomplextraits
hasbeenobserved.Encouragedbythisprogressthegenomewideassociationstudies(GWAS)of
humanagingandlongevityhavebeenperformed.Theresultsofthesestudiesweremuchless
impressive,however.StrongassociationsofgeneticvariantslinkedtoAPOE,FOXO3Aandtoseveral
othergeneswithhumanlifespanobservedinanumberofstudieswereaccompaniedbymany
associationsthathavenotreachedthelevelofgenomewidestatisticalsignificance.Mostresearch
findingssufferedfromthelackofreplicationinstudiesofindependentpopulations.Inthispaperwe
investigatereasonsthatmightberesponsibleforslowprogressingeneticanalysesofdataonaging
andlongevitytraits.Weshowedthatonesuchreasondealswiththefactthatbio‐demographic
aspectsofagingandlongevitytraitshavebeenignored.Thegeneticstructureofstudypopulation
getsmodifiedasaresultofmortalityselectionprocesswhichtakesplaceinanygenetically
heterogeneouspopulationswhensomegenesinfluencemortalityrisk.Suchmodificationaffectsthe
resultsofassociationstudies.Wediscussbenefitsofusingbio‐demographicconceptsandmodelsin
GWASofhumanagingandlongevity.Usingsimulateddata,andthentheFraminghamHeartStudy
dataweshowhowestimatesofgeneticassociationswithlifespancanbeimproved.Otherreasons
includingmultifactorialnatureofagingandlongevitytraits,highgeneticheterogeneityofthese
traits,pleiotropiceffectsofgeneticvariantsonmortalityrisksatdifferentageintervalsare
discussed.
P94
HeritabilityestimatesonHodgkinlymphoma:agenomicversus
populationbasedapproach
HaukeThomsen1,MiguelInaciodaSilvaFilho1,AstaFörsti1,MichaelFuchs2,ElkePoggevon
Strandmann2,PerHofmann3,StefanHerms3,JanSundquist4,AndreasEngert2,KariHemminki1
1GermanCancerResearchCenter(DKFZ),DivisionofMolecularGeneticEpidemiology,Heidelberg,69120,
Germany
2DepartmentofInternalMedicineI,UniversityHospitalofCologne,Cologne,50924,Germany
3InstituteofHumanGeneticsandDepartmentofGenomics,UniversityofBonn,53127,Germany
4StanfordPreventionResearchCenter,StanfordUniversitySchoolofMedicine,Stanford,94305,USA.
Genome‐wideassociationstudies(GWAS)haveidentifiedseveralsingle‐nucleotidepolymorphisms
influencingtheriskofHodgkinlymphoma(HL)anddemonstratedtheassociationofcommon
geneticvariationforthistypeofcancer.Suchevidenceforinheritedgeneticriskisalsoprovidedby
thefamilyhistoryandveryhighconcordancebetweenmonozygotictwins.However,littleisknown
aboutthegeneticandenvironmentalcontributions.Acommonmeasurefordescribingthe
phenotypicvariationduetogeneticsistheheritability.UsingGWASdataon906HLcasesby
consideringalltypedSNPssimultaneously,wehavecalculatedthatthecommonvarianceexplained
bySNPsaccountsformorethan35%ofthetotalvariationontheliabilityscaleinHL(95%
confidenceinterval6–62%).Thesefindingsaresupportedbysimilarheritabilityestimatesofabout
0.40(95%confidenceinterval0.17‐0.58)basedonSwedishpopulationdata.Ourestimatessupport
theunderlyingpolygenicbasisforsusceptibilitytoHL,andshowthatheritabilitybasedonthe
populationdataissomehowlargerthanforthegenomicdataduetothepossibilityofsomemissing
heritabilityintheGWASdata.BesidesthatthereisstillmajorevidenceformultiplelocicausingHL
onchromosomesotherthanchromosome6,whichneedtobedetected.Duetolimitedfindingsin
priorGWASitseemstobeworthtocheckformorelocicausingsusceptibilitytoHL
Categories: Association:Genome‐wide,Cancer,Case‐ControlStudies,Heritability,PopulationGenetics
P95
Areweabletoguidetreatmentchoicetoreduceantidepressant‐induced
sexualdysfunctioninmalesusinggenome‐widedatafromrandomised
controlledtrials?
AndrewACrawford1,SarahLewis1,KarenHodgson2,PeterMcGuffin2,DavidNutt3,TimJPeters4,
PhilipCowen5,MichaelCO'Donovan6,NicolaWiles1,GlynLewis7
1SchoolofSocialandCommunityMedicine,UniversityofBristol,Bristol
2MRCSocial,GeneticandDevelopmentalPsychiatryCentre,InstituteofPsychiatry,King'sCollegeLondon
3DepartmentofNeuropsychopharmacology,ImperialCollege,London
4SchoolofClinicalSciences,UniversityofBristol,Bristol
5DepartmentofPsychiatry,UniversityofOxford,WarnefordHospital,Oxford
6MRCCentreforNeuropsychiatricGeneticsandGenomics,SchoolofMedicine,CardiffUniversity,Cardiff
7DivisionofPsychiatry,UniversityCollegeLondon,London
Antidepressantsareeffectiveatreducingdepressivesymptomsbutarealsofrequentlyassociated
withincreasedsexualdysfunction.Treatmentemergentsexualdysfunctionisdrugspecific
(selectiveserotoninreuptakeinhibitor(SSRI)ornoradrenalinereuptakeinhibitor(NARI))andmay
begeneticallydetermined.Identifyinggeneticmarkersabletoguidetreatmentchoicewouldbe
clinicallyimportant.TheGENPODstudyrandomlyallocated601depressedindividualsto
citalopram(SSRI)orreboxetine(NARI).Analysiswasrestrictedtowhite,Europeanmenwithdata
onsexualdysfunction(n=105).Genome‐widedatawereanalysedusinglogisticregressioninan
additivegeneticmodel,withaninteractiontermbetweengenotypeanddrug.Replicationanalysis
useddatafromtheGENDEPstudy(n=202).Quantile‐quantileplotssuggestthatpopulation
stratificationwasgenerallywellcontrolled(lambda=1).Noassociationreachedagenome‐wide
levelofsignificance.GeneticvariantsnearthePTPRD(P=0.0001)andPCDH9(P=5.29x10‐5)genes
providedthestrongestevidenceofanassociationhowever,therewasnoevidenceinourreplication
cohort(P>0.1).Thelackofbiologicalplausibilityinouridentifiedgenescombinedwithalackof
evidenceinourreplicationstudyleadustoconcludethatlargertrialsarerequiredbefore
pharmacogeneticsmaybeabletoguideclinicalpracticeinthisarea.Theutilisationofdatafroma
randomisedcontrolledtrialandtheinclusionofaninteractionterminourregressionmodel
allowedustoidentifygeneticvariantswhoseassociationwithsexualdysfunctiondifferedby
antidepressant,whichareclinicallyimportant,butalsoincreasedourchancesofobtainingspurious
associations.
P96
AGENERALMETHODFORTESTINGGENETICASSOCIATIONWITHONEOR
MORETRAITS
ZenyFeng1,WilliamWLWong2
1DepartmentofMathematicsandStatistics,UniversityofGuelph
2LeslieDanFacultyofPharmacy,UniversityofToronto
Geneticassociationstudyisanessentialstepforfindinggeneticfactorsthatareassociatedwitha
complextrait.Manymethodshavebeenproposedforanalysingdatacollectedfromdifferentstudy
designs.Inthistalk,wewillpresentaverygeneralmethodthatbasedonthequasi‐likelihood
scoringapproachforanalysingdatacollectedfromabroadrangeofstudydesigns.Theproposed
methodcanalsobeusedtosimultaneouslytestonmultipletraits.Simulationstudiesandrealdata
analysiswillbeincludedtoshowtheperformanceoftheproposedmethod.
Categories: Association:Genome‐wide,MultivariatePhenotypes
P98
AGeneralizedSimilarityUtestforMultiple‐traitSequencingAssociation
Analyses
ChangshuaiWei1,QingLu1
1DepartmentofEpidemiologyandBiostatistics,MichiganStateUniversity
Sequencing‐basedstudiesareemergingasamajortoolforgeneticassociationresearchoncomplex
diseases.Thesestudiesposeagreatchallengetotraditionalstatisticalmethods(e.g.,singlevariant
analysis)duetothehigh‐dimensionalityofthedataandthelowfrequencyofthegeneticvariants.A
jointtesthasbeenshowntobemoresuitableforsequencingstudies;jointlytestingmultiple
variantsincreasesthepowerandreducesthedimensionality.Meanwhile,thereisagrowingneed
forstatisticalmethodsthataredistribution‐freeandthatcanhandlemultiplephenotypes.Inthis
paper,weproposeageneralizedsimilarityUtest,referredtoasGSU.GSUfirstsummarizesthe
geneticinformationandmultipletraitsintoageneticsimilarityandatraitsimilarity,andthen
combinesthetwosimilaritiesintheframeworkofaweightedUstatistic.Wederivedtheasymptotic
distributionofGSUunderanullhypothesis,soastoefficientlycalculatethesignificancelevel.We
alsostudiedtheasymptoticbehaviorofGSUunderalternatives,andprovidedsamplesizeand
powercalculationsforthestudydesign.ToevaluatetheperformanceofGSU,weconducted
extensivesimulationstudiesandcomparedthemwiththeexistingmethods.Throughsimulation,we
foundthatGSUhadanadvantageoverexistingmethodsintermsofpowercomparisonsandits
robustnesstotraitdistribution.Moreover,GSUiscomputationallymoreefficientthanexisting
methods.Finally,weappliedGSUtosequencingdatafromtheDallasHeartStudy,identifying4
genesjointlyassociatedwith5metabolic‐relatedtraits.
Categories: Association:UnrelatedCases‐Controls,Case‐ControlStudies,MultilocusAnalysis,
MultivariatePhenotypes,SequencingData
P99
ModelingX‐chromosomedatainRandomForestGeneticAnalysis
JoannaMBiernacka1,GregoryJenkins1,StaceyJWinham1
1MayoClinic
TheXchromosomeisroutinelyexcludedfromgenome‐wideassociationstudies.RandomForests
(RF)havebeenproposedforgeneticanalysisinvolvingmanyvariants.Weillustratethatfortraits
associatedwithsex,RFanalysisyieldsbiasedresultsforXchromosomeSNPs,andproposethree
extensionsofRFtomodelXSNPs,basedon(1)theprincipleofXchromosomeinactivation(XCI),(2)
stratificationbysex,and(3)incorporationofsexasavariableinRF.Wecomparetheperformance
oftheseapproachestotraditionalRFusingsimulationsandanalysisofdatafromtheStudyof
Addiction:GenesandEnvironment(SAGE).ComparisonoftheSAGEdataresultsforautosomalvs.X
SNPsshowsthattraditionalRFranksXSNPstoohigh,whereasthethreenewapproachesranktheX
SNPssimilartoautosomalSNPs.ToevaluatethealternativeapproachestoincorporatingX‐SNPdata
inRF,weinvestigatevariableimportance(VI)measuresforautosomalandXSNPsinsimulateddata
withandwithoutXSNPeffects.Weperformsimulationsundervaryingdegreesofsex‐trait
association,case/controlratios,andpatternsoflinkagedisequilibrium.Allmethodscorrectly
estimateVIifsexisnotassociatedwiththetrait,butwhensexisassociatedwiththetrait,
traditionalRFleadstoinflatedVIfortheXchromosome.IncorporatingsexinRFdoesnotproperly
correctthisbias,whereasthemethodsbasedonXCIandstratifiedRFdonotinflatetheVIofXSNPs.
Thus,weconcludethatifsexisnotassociatedwiththetraitinthesample,regularRFmaybeused
toanalyzeXSNPdata.Otherwise,eitherstratificationoftheforestorextensionbasedonXCIshould
beusedtoavoidoverestimationofXSNPimportance.Futureinvestigationwillcomparethepower
ofthesetwomethods.
Categories: Association:UnrelatedCases‐Controls,Case‐ControlStudies,DataMining,Machine
LearningTools
P100
EmpiricalBayesScanStatisticsforDetectingClustersofDiseaseRisk
VariantsinGeneticStudies,withApplicationstoCNVsinAutism
IulianaIonita‐Laza1,KennethMcCallum1
1ColumbiaUniversity
Recentdevelopmentsofhigh‐throughputsequencingtechnologiesofferanunprecedenteddetailed
viewofthegeneticvariationinvarioushumanpopulations,andpromisetoleadtosignificant
progressinunderstandingthegeneticbasisofcomplexdiseases.Despitethistremendousadvance
indatageneration,itremainsverychallengingtoanalyzeandinterpretthesedataduetotheir
sparseandhigh‐dimensionalnature.HereweproposeseveralempiricalBayesscanstatisticsto
identifygenomicregionssignificantlyenrichedwithrarediseaseriskvariants.Weshowthatthe
empiricalBayesmethodologycanbemorepowerfulthanexistingmethodsespeciallysointhe
presenceofmanynon‐diseaseriskvariants,andinsituationswhenthereisamixtureofriskand
protectivevariants.Furthermore,theempiricalBayesapproachhasgreaterflexibilityto
accommodatecovariatessuchasfunctionalpredictionscoresandadditionalbiomarkers.Weapply
theproposedmethodstoawholeexome‐sequencingstudyonautismspectrumdisordersand
identifyseveralnewgenesthatresideincopynumbervariableregionsassociatedwithautism.In
particular,genesSYNGAP1andRNF135arebothstrongcandidategenesforautismandhavebeen
identifiedbytheproposedmethods.
Categories: Association:UnrelatedCases‐Controls,SequencingData
P101
Finemappingofchromosome5p15.33regionforlungcancer
susceptibilitybasedonatargeteddeepsequencingandcustomAxiom
array
LindaKachuri1,ChristopherIAmos2,LoicLeMarchand3,ShelleyTworoger4,GeoffreyLiu5,JamesD
McKay6,PaulBrennan6,JohnKField7,JohnRMcLaughlin8,YafangLi2,RobertEDenroche9,PhilipC
Zuzarte9,JohnMcPherson9,RayjeanJHung1
1Lunenfeld‐TanenbaumResearchInstituteofMountSinaiHospital,Toronto,ON,Canada
2GeiselSchoolofMedicine,DartmouthCollege,Lebanon,NH,USA
3UniversityofHawaii,Honolulu,HI,USAShelley
4HarvardSchoolofPublicHealth,Boston,MA,USA
5OntarioCancerInstitute,PrincessMargaretCancerCenter,Toronto,ON,Canada
6InternationalAgencyforResearchonCancer,Lyon,France
7InstituteofTranslationalMedicine,UniversityofLiverpool,Liverpool,UK
8PublicHealthOntario,Toronto,ON,Canada
9GenomeTechnologies,OntarioInstituteforCancerResearch,Toronto,ON,Canada
Background:Genome‐wideassociationstudieshaveconsistentlylinkedsinglenucleotide
polymorphisms(SNPs)inch5p15.33withincreasedlungcancerrisk.Thisregioncontainstwo
knowncancersusceptibilitygenes:telomerasereversetranscriptase(TERT)andcleftlipandpalate
transmembrane1‐like(CLPTM1L),however,thecausalmechanismsunderlyingtheseriskvariants
havenotbeenfullyelucidated.Methods:Wecarriedoutafinemappingof5p15.33firstbydeep
sequencing288lungcancercase‐controlpairs,andsubsequentlygenotyping4608SNPs(1125de
novovariants:953SNPs,172indelsnotpreviouslydescribed)usingacustomAffymetrixAxiom
arrayin3063casesand2940controlsofEuropeanancestryfrom5studies:MSH‐PMH,EPIC,MEC,
LLPC,HPFS&NHS.Oddsratios(OR)adjustedforage,sexandcigarettepack‐yearswereestimated
usinglogisticregression.Sequencekernelassociationtests(SKAT)wereusedtolocalizetheeffects
ofrarevariants.Results:17SNPsmetthemultipletestingcorrectedthreshold(p<4.1×10‐4).Of
these,twonewlyidentifiedvariantswerestronglyassociatedwithlungcancerriskafter
conditioningontheeffectsofknownriskvariants:ch5:1253720(OR:0.25,p=6.6×10‐6)locatedin
theTERTexon,andch5:1384599(OR:0.03,p=3.1×10‐4)downstreamofCLPTM1L.13ofthe17
significantSNPswerelocatedinCLPTM1L.TheSKATanalysispointstoriskvariantswithinthe
TERTexon(p=8.8×10‐4),downstreamofCPTM1L(p=1.5×10‐4)andmicroRNA4457(p=4.7×10‐4).
Conclusions:Inthisstudyweidentifiedseveralnovelvariantsthatwereindependentlyand
significantlyassociatedwithlungcancerrisk.Ourfindingsrefinedtheassociationbetweenthe
TERT/CLPTM1Lregionandlungcancerrisk.
Categories: Association:UnrelatedCases‐Controls,Cancer,FineMapping,SequencingData
P102
Geneticvariantsininflammation‐relatedgenesandinteractionwith
NSAIDuseoncolorectalcancerriskandprognosis
YesildaBalavarca1,NinaHabermann1,DominiqueScherer1,KatharinaBuck1,PetraSeibold2,Katja
Butterbach3,BarbaraBurwinkel4,KatrinPfuetze4,MichaelHoffmeister3,ElisabethKap2
1DivisionofPreventiveOncology,NationalCenterforTumorDiseases(NCT/DKFZ),Heidelberg,Germany
2DivisionofCancerEpidemiology,UnitofGeneticEpidemiology,GermanCancerResearchCenter(DKFZ),
Heidelberg,Germany
3DivisionofClinicalEpidemiologyandAgingResearch,GermanCancerResearchCenter(DKFZ),Heidelberg,
Germany
4MolecularEpidemiology,GermanCancerResearchCenter(DKFZ),Heidelberg,Germany
Inflammationhasbeenshowntocontributetocolorectalcarcinogenesis.Non‐steroidalanti‐
inflammatorydrugs(NSAIDs)areassociatedwithreducedinflammation.Thus,westudied
associationofinflammation‐relatedgenesandtheirinteractionwithNSAIDuseregardingcolorectal
cancer(CRC)riskandsurvival.Weanalyzed15genes(169SNPs)of1756CRCpatientsand1781
controlsenrolledinacase‐controlstudywithfollow‐upofpatients(DACHS).CRCriskwasassessed
bymultivariableunconditionallogisticregressionandoverallsurvivalbymultivariableCox
regressionmodels.Pvaluesfornon‐candidateSNPswereadjustedformultipletesting(padj.)CRP
(rs1205,p=0.04;rs1800947,p=0.02)andPTGS1(rs10513402,padj.=0.01)variantswere
associatedwithincreasedCRCrisk.SubjectswiththevariantalleleinCRP(rs1800947,p=0.004)
andinPTGIS(rs477627,p=0.04),respectively,showedlowerCRCriskwiththeuseofNSAIDs.After
5‐yearsfollowup,variantsinPTGS1wereassociatedwithpooreroverallsurvival(rs1330344,p
adj.=0.02andrs3119773,padj.=0.04).NSAIDusewasassociatedwithimprovedoverallsurvivalof
patientswiththevariantalleleinCCL2(rs3760396,p.adj=0.01)andwithdecreasedoverallsurvival
ofpatientswiththevariantalleleinIL23R(rs12041056,p=0.008).InpatientswithdiseasestageI‐
III,thosewithvariantsinIL18(rs1293344,padj.=0.01)andinIL23R(rs10889665,padj.=0.02)
showedimprovedoverallsurvival.Weshowedthatgeneticvariationsininflammation‐related
genesandtheirinteractionswithNSAIDsareassociatedwithCRCriskandsurvival.This
informationmayaidintailoringpreventionstrategiestosubjectswhowillbenefitmostfromNSAID
use.
Categories: Association:UnrelatedCases‐Controls,Cancer,Gene‐EnvironmentInteraction,
P103
AssociationanalysisofexomechipdataofPolycysticOvarySyndromein
EstonianBiobank
ReedikMägi1,AndrewPMorris2,TKaraderi3,TriinTriinLaisk‐Podar4,TriinTammiste4,Andres
Metspalu1,AndresSalumets4,CeciliaMLindgren5
2DepartmentofBiostatistics,UniversityofLiverpool,Liverpool,UK
3WellcomeTrustCentreforHumanGenetics,UniversityofOxford,Oxford,UK
4DepartmentofObstetricsandGynaecology,UniversityofTartu,Tartu,Estonia
5BroadInstituteoftheMassachusettsInstituteofTechnologyandHarvardUniversity,Cambridge,MA,USA
Polycysticovarysyndrome(PCOS)isacommonmultifactorialdiseaseaffectingupto10%of
womenofreproductiveage.Toinvestigatethecontributionofpotentiallycausalcodingvariantsto
PCOS,wehavegenotyped167casesand711populationcontrols(363females)fromtheEstonian
BiobankwiththeIlluminaexomearray.Weconductedsinglevariantandburdentestsofassociation
usingSKAT‐Owithingenesfor(i)lossoffunction(LOF)and(ii)rarenon‐synonomous(NS)variants
withminorallelefrequency(MAF)<1%).Theassociationanalyseswereadjustedforfirsttwo
principalcomponentstoaccountforthepopulationstratification.Intheautosomalanalysis,both
maleandfemalesampleswereusedinthecontrolgroupbutintheXchromosomeanalysis,only
femalesampleswereused.Altogether55,345polymorphicvariantsweresuccessfullytestedin
singlevariantanalysis.Itrevealedonemissensevariantwhichwasshowingexome‐wideevidence
ofassociation(p<5x10‐7,Bonferronicorrectionfor100,000variants):exm233350inthenebulin
codingNEBgene(p=4.9x10‐9,MAF=0.05%).MutationsinNEBhavepreviouslybeenassociated
withmyopathyandmusclestructure.Noneoftheassociationswerestatisticallysignificantinthe
gene‐basedtestsaftermultipletestingcorrectionfor20,000genes.Thestrongestassociationscame
fromaggregatingnon‐synonymousrarevariantswithinPOLK(p=4.3x10‐5)andPELI3(7.3x10‐5)
gene,whichareDNAreplicationandimmuneresponserelatedgenes.Ourstudysuggeststhatrare
variantscancontributetothegeneticcomponentofPCOS,butcannotexplainpreviouslyreported
associationsignalsinestablishedGWASloci.
Categories: Association:UnrelatedCases‐Controls,Case‐ControlStudies
P104
AMODELFORCO‐SEGREGATIONOFCRYPTORCHIDISMANDTESTIS
CANCERINFAMILIES
DuncanCThomas1,VictoriaKCortessis1
1UniversityofSouthernCalifornia
Testiculargermcelltumors(TGCT)andcryptorchidism(CO)arehighlyfamilial,butlociidentified
bygenome‐wideassociationstudiesexplainonly15‐22%ofTGCTheritability.Tounderstand
segregationofthetraitsanddependencyofTGCTonCO,wedevelopedanovelstatisticalmodel
incorporatingmajorgenes,polygenes,andnongeneticfrailtiesaccountingfordependencebetween
testes,foreachtraitandthetransitionfromCOtoTCGT.From17,844TCGTcasesintheCalifornia
CancerRegistry,weobtainedinformedconsentandpersonalandfamilyhistoryfrom5,702(17,844
familymembers),andextendedpedigreesfor697ofthosecasesreportingbilateralTGCT,CO,or
familyhistoryofeithertrait(23,143members).Adjustingforthiscomplexascertainment,wefound
strongevidenceforpolygeniceffectsforCOandTCGTandamajorgenemodifyingtheeffectofCO
onTCGTrisk.Genotypesfor9TGCTriskvariantsin1,639membersof527familieswereusedinan
extendedmodelincorporatingmultiplegenesassumedtobeinLDwiththeSNPsandtosegregate
withoutrecombination.ThisrevealedsignificantassociationsofCOwithTERTandCENPE,baseline
TGCTriskwithKITLGandUCK2,andsuggestiveevidencethatTERTmodifiestheeffectofCOon
TGCT.Thesesupportageneticbasisforfamilialaggregationofbothtraitsanddependencybetween
traits.Themodelcanaddressotherprecursors(e.g.polypsforcolorectalcancer,mammographic
densityforbreastcancer).
Categories: Ascertainment,Association:CandidateGenes,Association:Family‐based,BayesianAnalysis,
Cancer,FamilialAggregationandSegregationAnalysis,LinkageandAssociation,MarkovChainMonte
CarloMethods
P105
Jointanalysisofsecondaryphenotypes:anapplicationinfamilystudies
RenaudRTissier1,RoulaSTsonaka1,JeanineJHouwing‐Duistermaat1
1LUMC,TheNetherlands
Acase‐controldesignistypicallyusedinordertotestassociationsbetweenthecase‐controlstatus
(primaryphenotype)andgeneticvariants.Inadditiontothisprimaryphenotypesecondarytraits
areavailableandassociationsarestudiedbetweenthegeneticvariantsandthesesecondarytraits.
However,whenanalyzingthesephenotypesthecase‐controldesignhastobetakenintoaccount
especiallywhenthemarkertestedisassociatedwiththeprimaryphenotypesorwhenthereis
correlationbetweentheprimaryandsecondaryphenotype.Methodsareavailableforsecondary
phenotypeanalysisincase‐controlstudies.Thesemethodsarenotdirectlyapplicabletomore
complexdesigns,suchasmultiplecasesfamilystudies.Hereapropersecondaryanalysisis
complicatedbythebiasedsamplingdesign,thewithinfamiliescorrelationsandthemixedtypeof
outcomes:binaryprimaryphenotypeandcontinuoussecondaryphenotypes.Weproposeabias
correctionapproachforsecondaryphenotypeanalysisinfamilystudieswhichallowsinvestigation
ofgeneticeffectsacrossmultiplesecondarytraits.Weadopttheretrospectivelikelihoodmethodto
correctforascertainmentofthefamiliesanduseacorrelatedprobitmodeltomodeljointlythe
mixedtypeprimaryandsecondaryphenotypes.Theestimatesoftheparameterscanbepooledwith
resultsfromstudiescomprisingrandomlyselectedsubjectsbystandardmetaanalysistools.We
studiedtheperformanceviasimulationsandestimatedtheeffectsofseveralSNPsonatriglyceride
availableintheLeidenLongevityStudy,acasefamily‐controlstudy.Weconcludethattheuseofan
ad‐hocwillleadtobiasespeciallyincasetheSNPisassociatedwiththeprimaryphenotype
Categories: Ascertainment,Association:CandidateGenes,Association:Family‐based,Association:
UnrelatedCases‐Controls
P106
Predictionofimprintedgenesbasedonthegenome‐widemethylation
analysis
NataliaTšernikova1,NeemeTõnisson2,KaieLokk2,AndresSalumets3,AndresMetspalu1,Reedik
Mägi4
1DepartmentofBiotechnology,InstituteofMolecularandCellBiology,UniversityofTartu,Tartu,Estoniaand
EstonianGenomeCenter,UniversityofTartu,Tartu,Estonia
2DepartmentofBiotechnology,InstituteofMolecularandCellBiology,UniversityofTartu,Tartu,Estonia
3CompetenceCentreonReproductiveMedicineandBiology,Tartu,Estonia;DepartmentofObstetricsand
Gynecology,UniversityofTartu,Tartu,EstoniaandInstituteofBiomedicine,UniversityofTartu,Tartu,
Estonia
4EstonianGenomeCenter,UniversityofTartu,Tartu,Estonia;
Genomicimprintingisanepigeneticgene‐markingphenomenonthatisestablishedingermline.Our
hypothesisisthatimprintedgenescanbepredictedbythemethylationlevel.Weexpectsemi‐
methylationinimprintedgenes.InordertoprovethishypothesisweanalysedtheDNAmethylation
inwell‐knownimprintedgenesacrossthetissuepanelfromthesameindividuals.17tissuesfrom4
individualswerecollectedduringautopsy.DNAmethylationanalysisofthetotal72tissuesamples
wasperformedwiththeIlluminaInfiniumHumanMethylation450BeadChip.WeusedLevene’stest
forcomparisonofknownimprintedgeneswiththerestofthegenescapturedby450Kmethylation
array.Asaresult,allimprintedgenes(n=92)demonstratedlessvariabilityinthemethylationlevel
(p<0.01)acrossalltissues.WealsovisualizedCpGpatternsofknownimprintedgenesacrossall
tissues.EachCpGwasannotatedtoitsexactlocationinthegenomeinexon,genebodyorUTR
region.VisualizedCpGpatternsalsoconfirmedtissue‐specificnatureofimprintedgenes.For
example,gallbladdershowsmediummethylationofKCNQ1DNgene,whileinisciaticnervetheCpG
sitesarenotmethylated.Usingthismappingmethod,wenarroweddownthelistofpotential
candidategenesto3000.Wefoundthatsomegenesmeetthecriteriaforcandidateimprintedgenes
inallsomatictissues,whileothergenesmeetthosecriteriaonlyinsomeofthetissues.Asthenext
stepweareusingtheRNAseqdatatofurthernarrowdownthelistofcandidategenes.Ourmethod
canberegardedasatooltoidentifythetissuespecificityofthealreadyestablishedimprintedgenes
aswellastodiscovernewimprintedgenesacrossthewholehumangenome.
Categories: Bioinformatics,EpigeneticData,Epigenetics
P107
AddictionandMentalHealthGenesformGenomicHotspotswith
DrugableTargets.
LatifaFJackson1,AydinTozeren1
1DrexelUniversity
Dopamine,alcoholandopiateaddictionarewellcharacterizedco‐morbiditieswithdepression,
bipolarandschizophreniadisorders.Whileeachofthesedisordersareknowntohaveastrong
geneticbasis,thereislittlesystematicinformationthataddressesthegeneticintersectionsofthese
co‐morbiddisorders.Wecanharnesscuratedgenesetsderivedfromsinglegeneandgenomewide
associationstudiestoidentifythegenomicregionsdisproportionatelyparticipatinginaddictionand
mentalhealthdisorders.Opiate,dopamineandalcoholaddictiondisordergenesetsandmental
healthgenesets(schizophrenia,depressionandbipolardisorder)obtainedfromNationalCenterfor
BiotechnologyInformationGene,werecombined,thenprojectedontothegenome,andtheregions
ofinterestwereannotatedwithgeneontologycategoriesandcellularpathways.functional
annotationsamongtheresultingaddictionandmentalhealthgenomichotspotregionsforma
bioinformaticsportraitofthegeneticintersectionslikelytocontributetoobservedco‐morbidities.
Weidentifyeightgenomichotspots,withanoverabundanceofaddictionandmentalhealthgenes
(p<0.005).Hotspotgenesandtheirco‐locatedcandidatecounterpartsareinvolvedinsignificant
coreneurologicalfunctions(p=0.05):neurologicaltransmission,responsestoorganicsubstances
andcell‐cellsignaling.Wefurtherannotatedallhotspotgenesfortheirassociateddrugbindingsites
toidentifywhetherthesebindingsiteswerecandidatesfortherapeuticintervention.Wefound16
drugbindingsites,whichsortedintofourthematicclasses:anillicitdrugbindingsite,fourmental
healthdrugsites,threeimmuneresponsesites,andfivecancerbindingsiteswithstrongaddiction
ormentalhealthcounter‐indications.Ouranalysesdemonstratetheutilityofconsideringahotspot
approachinidentifyinggenomicregionscontributingtotheintersectionofaddictionandmental
healthandprovidegenecandidatesforpotentialdrugtargets.Keywords:Bioinformatics,Addiction,
Schizophrenia,BipolarDisorder,Depression,Genome,Alcohol,CausalInference,CandidateGenes
Categories: Bioinformatics,Gene‐GeneInteraction,GenomicVariation
P108
Recurrentsharedrarevariantsin9genesdetectedbywholeexome
sequencingofmultiplexoralcleftsfamilies
JoanEBailey‐Wilson1,EmilyRHolzinger1,QingLi1,MargaretMParker2,JacquelineBHetmanski2,
MaryLMarazita3,LLeighField4,AjitRay5,ElisabethMangold6,MarkusMNöthen6
1ComputationalandStatisticalGenomicsBranch,NationalHumanGenomeResearchInstitute,National
InstitutesofHealth,Baltimore,MD,USA
2DepartmentofEpidemiology,BloombergSchoolofPublicHealth,JohnsHopkinsUniversity,Baltimore,MD,
USA
3CenterforCraniofacialandDentalGenetics,DepartmentofOralBiology,UniversityofPittsburgh,Pittsburgh,
PAUSA
4Emeritus,UniversityofBritishColumbia,Vancouver,BCCanada
5Emeritus,UniversityofToronto,Toronto,Ontario,Canada
6InstituteofHumanGenetics,UniversityofBonn,Bonn,Germany
Non‐syndromiccleftlipwith/withoutcleftpalate(CL/P)isacomplextrait.Genome‐wide
associationstudies(GWAS)haveidentifiedseveralgeneticriskfactorsforCL/Pandrecentlywe
identifiedanovel,potentiallydamagingvariantinCDH1inoneIndianmultiplexCL/Pfamily.Here,
weusedwholeexomesequence(WES)dataon2or3related(2oormoredistant)affected
individualsperfamilytoidentifygenescontainingsharedrarevariants(RV).Fifty‐fivefamiliesof
Indian(12),Filipino(11),German(19),Syrian(10),European‐American(1)andAsian(2)descent
containing114individuals,4duplicatecontrolsand2unrelatedCEPHHAPMAPcontrolswere
sequencedontheIlluminaHi‐Seq2500andprocessedthroughGATK.Ingenuity‘VariantAnalysis’
wasusedtoidentifyRVssharedunderarecessivemodelbyallsequencedaffectedindividualsina
family.Geneswheresuchsharingwasobservedinthesamegeneinatleast2separatemultiplex
families(differentRVsperfamily)wereconsideredpotentiallyrelatedtoCL/P.Afterfilteringbased
onvariantqualityandfrequency(MAF<0.05),weidentified9genesexhibitingrecessiveRVsharing
inallaffectedindividualsinatleasttwofamilies:ARHGEF12,CCT4,HSD3B7,MAN1B1,RREB1,
SNRPC,STARD9,ZDHHC11,andZNF835.TheseRVsarenotpresentineitherofthesequenced
HapMapcontrols.Follow‐upwillincludeSangersequencingandgenotypingoftheseRVsinother
affectedandunaffectedindividualsinthesesamefamiliestodetermineiftheRVssegregatewith
disease.
Categories: Bioinformatics,DataMining,GenomicVariation,SequencingData
P109
Evaluationofvariantcallingfromthousandsoflowpasswholegenome
sequencing(WGS)datausingGATKhaplotypecaller
HuaLing1,KurtHetrick1,PengZhang1,ElizabethPugh1,JaneRomm1,KimberlyDoheny1
1CenterforInheritedDiseasesResearch(CIDR),JohnsHopkinsUniversity
LowpassWGS(~2‐8x)onlargenumbersofsampleshasbecomeanattractivestrategyingenetic
studiesofcomplextraits.Giventhesameamountofsequenceyield,itmayprovidemorepowerin
detectingdiseaseassociatedvariantsthandeepsequencing(30x)asmallernumberofsamples.It
canalsobeusedtobuildareferencepanelforimputingadditionalsamplestofurtherboostpower.
RecentadvancesinGATKenableustodojointvariantcallingandanalysisonmultipleWESsamples
usingHaplotypeCaller(HC)thatwascomputationallyprohibitive.HCisdesirableoverUnified
Genotyper(UG)notonlybecauseofitshigheraccuracyinvariantcalling(especiallyforINDELs),but
itoffersgreaterflexibilitybyallowingforaddinginmoresamplesatalaterstagewithoutre‐
processingthecohort.Toevaluatethefeasibilityandperformanceofcallingthousandsoflowpass
WGSsamplesundercurrenthardwareandsoftwaresupports,weused2,535lowpassWGBAMfiles
fromthe1KGPforchr11.WegeneratedgenomicVCFfilesforeachindividualsamplewithHC,and
createdjointcallsusingGenotypeGVCFsfollowedbyvariantfilteringwithVQSR.Themeancoverage
persamplerangedfrom2.8to38(medianof6.67),with70%sampleshavemeancoveragebelow
8x.Morethan96%and85%baseshavedepthgreaterthan2Xand4X,respectively.ForNA12878
(meancoverageof4.91),lowpassWGSmade~75%ofSNVsonchr11thatarecalledby
GenomeInABottle(v2.18).Bycomparingthistoarraydataand30xWGSgeneratedonsitefora
subsetofsamplesandreviewedusingIGV,wecancharacterizesensitivityandconcordanceat
differentlevelsofMAFandsequencingdepthforbothSNVsandINDELs.
Categories: Bioinformatics,DataQuality,GenomicVariation,SequencingData
P110
IntegrationoffMRIandSNPsindicatedpotentialbiomarkersfor
Schizophreniadiagnosis
HongbaoCao1,Yu‐PingWang2,VinceCalhoun3,YinYaoShugart1
1NationalInstituteofHealth
2TulaneUniversity
3UniversityofNewMexico
Integrativeanalysisofmultipledatatypescantakeadvantageoftheircomplementaryinformation
andthereforemayprovidegreaterpowertoidentifypotentialbiomarkers.However,duetothe
diversityofthedatamodality,dataintegrationischallenging.Hereweaddressthedataintegration
problembydevelopingageneralizedsparsemodel(GSM)usingweightingfactorstointegrate
multi‐modalitydataforbiomarkerselection.Toprovethefeasibility,weappliedtheGSMmodeltoa
jointanalysisoftwotypesofschizophreniadatasets:759075SNPsand153594functionalmagnetic
resonanceimaging(fMRI)voxelsin208subjects(92cases/116controls).Tosolvethissmall‐
sample‐large‐variableproblem,wedevelopedanovelsparserepresentationbasedvariable
selection(SRVS)algorithm,aimingtoidentifybiomarkersassociatedwithschizophrenia.To
validatetheeffectivenessoftheselectedvariables,weperformedmultivariateclassification
followedbyaten‐foldcrossvalidation.ResultsshowedthatourproposedSRVSmethodcanbeused
toidentifynovelbiomarkersandofferstrongercapabilityindistinguishingschizophreniapatients
fromhealthycontrols.Moreover,betterclassificationratioswereachievedusingbiomarkersfrom
bothtypesofdata,suggestingtheimportanceofintegrativeanalysis.Especially,withnormbased
penalty,ourSRVSmethodgeneratedhighestclassificationaccuracyindiscriminatingschizophrenia
patientsfromhealthycontrols.Thissuggeststhatnormmaybethebestchoiceaspenalizationterm
fortheproposedSRVSmethod.Furtherbiologicalexperimentalworkisneededtovalidatethe
biomarkersidentifiedinthepaper.
Categories: Bioinformatics,Case‐ControlStudies,DataIntegration
P111
EWAStoGxE:Arobuststrategyfordetectinggene‐environment
interactionmodelsforage‐relatedcataract
MollyAHall1,JohnRWallace1,SarahAPendergrass1,RichardBerg2,TerrieKitchner2,PeggyPeissig2,
MurrayBrilliant2,CatherineAMcCarty3,MarylynDRitchie1
1CenterforSystemsGenomics,ThePennsylvaniaStateUniversity,UniversityPark,PA
2MarshfieldClinic,MarshfieldWI
3EssentiaRuralHealth,Duluth,MN
Gene‐environmentinteractions(GxE)areessentialtoelucidatingthenatureofcomplextraits,but
computationaldemandsandmultipletestingmakeuncoveringtheseinteractionsdifficult.We
addressthisusinganenvironmentwideassociationstudy(EWAS)toidentifyputative
environmentalfactorsinahigh‐throughputmannerfollowedbyatestforGxEwithgenome‐wide
SNPsforassociationwithcataract.WeperformedadietaryEWASbyevaluating57dietary
exposuresfromaDietaryHistoryQuestionnaireusinglogisticregression,adjustedforage,sex,and
type2diabetes(T2D)in2,629samples(932controls,1,697cases)ofEuropeandescentfromthe
MarshfieldClinicPersonalizedMedicineResearchProject,partoftheElectronicMedicalRecords&
Genomics(eMERGE)Network.Sevendietarymeasureswerepredictiveofcataract(p‐value<0.05);
amonounsaturatedomega‐9fattyacid,erucicacid(FA22:1)(p=5.5×10‐4)passedourBonferroni
correctedp‐valuethreshold.WethentestedFA22:1forGxEusing498,829SNPsinasubsetof
samplesforwhomgeneticdatawasavailable(831controls,1,511cases)usinglogisticregression
adjustedforage,sex,andT2Dstatus.TwentySNP‐FA22:1modelsweresignificant(p<1.0×10‐4).
ThemostsignificantGxEmodelwasFA22:1andrs726712,anintronicSNPinLPP(p=2.9×10‐5).
Theerucicacid‐cataractassociationisnovel;althoughtwopolyunsaturatedfattyacidshavebeen
foundincataractoushumanlenses.LPPencodesaproteininvolvedincell‐celladhesion,aprocess
withmultiplepublishedassociationswithcataract.ThesefindingsindicatetheroleofGxEin
susceptibilitytocataractanddemonstratetheutilityofEWASforinvestigatingtheGxEinterplayof
complexdiseases.
Categories: Bioinformatics,Case‐ControlStudies,Gene‐EnvironmentInteraction
P112
RNA‐seqanalysisoflungadenocarcinomarevealsdifferentialgene
expressioninnonsmokerandsmokerpatients
YafangLi1,XiangjunXiao1,ChristopherIAmos1
1Dartmouthcollege
Lungadenocarcinomaisacomplexdiseasethatcausedbybothgeneticandenvironmentaleffect.
TheRNA‐seqtechnologyprovidesusapowerfultoolfortranscriptomeanalysisoflungcancer.In
thisstudy,weusedRBioconductoredgeRtoanalyzeRNA‐seqfrompairednormalandtumortissue
in34nonsmokerand40smokerpatientswithlungadenocarcinoma(GEO:GSE40419).Wedivided
thesamplesintopilotandreplicationstudyforeachgroup,andthereishighconsistencebetween
theresultsfromreplicationandpilotstudies.Thegenedifferentialexpressionanalysisidentified
179genesthatshoweddifferentialexpressiononlyintumorsfromnonsmokerpatients;780genes
thataredifferentiallyexpressedinbothsmokerandnonsmokertumortissueversusnormaltissue;
and1869genesthatexclusivelyvariedintumortissuefromsmokerpatientsversusnormaltissue.
77%and59%oftheidentifiedgenesaredownregulatedinnonsmokerandsmokergroups,
respectively.Amongthecommongenes,thegenestendtohavealargerlogFCchangeinsmoker
patientsthannonsmokerpatients.Thesmokerandnonsmokerpatientspecificgeneswithlarge
logFCarealsoidentifiedinouranalysis.Ourstudyprovidesasystematicanalysisofwholegenome
genedifferentialexpression.Itprovidestargetgenesforsubsequentbiologicalstudiestodecipher
theaberrationsthatarepresentinlungadenocarcinoma.
Categories: Bioinformatics,Cancer,Case‐ControlStudies,GeneExpressionPatterns,SequencingData
P113
UsingrandomforeststoidentifygeneticlinksbetweenAlzheimer’s
diseaseandtype2diabetes
BurcuFDarst1,ChenYao2,RebeccaLKoscik3,BarbaraBBendlin4,BrucePHermann3,AsenathLa
Rue3,SterlingCJohnson5,MarkASager3,CorinneDEngelman1
1DepartmentofPopulationHealthSciences,UniversityofWisconsinSchoolofMedicineandPublicHealth,
Madison,WI,USA
2DepartmentofDairyScience,UniversityofWisconsin,Madison,WI,USA
3Alzheimer'sDiseasesResearchCenter,UniversityofWisconsinSchoolofMedicineandPublicHealth,
Madison,WI,USA;WisconsinAlzheimer'sInstitute,UniversityofWisconsinSchoolofMedicineandPublic
Health,Madison,WI,USA
Madison,WI,USA;GeriatricResearchEducationandClinicalCenter,Wm.S.MiddletonMemorialVAHospital,
Madison,WI,USA
Madison,WI,USA;WisconsinAlzheimer'sInstitute,UniversityofWisconsinSchoolofMedicineandPublic
Health,Madison,WI,USA;GeriatricResearchE
Increasingevidencesuggeststhattype2diabetes(T2D)isariskfactorforAlzheimer’sdisease(AD),
butthegeneticmechanismlinkingtheseconditionsisunknown.Usingrandomforests(RF),we
investigatedwhetherinteractionsbetweensinglenucleotidepolymorphisms(SNPs)inapathway
linkedtobothADandT2D,andriskfactorsforT2Dinfluencecognitioninacohortofmiddle‐aged
adultsenrichedforaparentalhistoryofAD.Weanalyzedasampleof836participantsfromthe
WisconsinRegistryforAlzheimer’sPreventionwithdataonpredictorsofT2D,30SNPsinthe
SORL1andSORCS1genes,and4cognitivefactors.ThesevariableswereinputintoRF,amachine‐
learningalgorithmthatcalculatesimportancescoresbasedonthevarianceexplainedbyeach
variableinamodelwhileallowingforinteractions.BecauseRFdoesnotspecificallyidentify
interactingvariables,weusedanovelapproachthatidentifiesinteractionsbydetermininghow
oftenapairofvariablesdescendstogetherinaRF.Rs7907690inSORCS1,andrs2282649and
rs1010159inSORL1,appearedinthetop25descendantpairsforall4cognitivefactors,frequently
pairedwithwaist‐hipratio,HOMA‐IR(ameasureofinsulinresistance),age,andphysicalactivity.
Manyoftheinteractionsidentifiedwithdescendantpairsconsistedofdiscordantlyrankedpairs,
withonevariablehavingahighimportancescoreandtheotherhavingalowimportancescore.
TheseresultssuggestthatinteractionsbetweenSNPsassociatedwithADandT2Dandriskfactors
forT2DmaycontributetotherelationshipbetweenADandT2Dandthatthedescendantpair
methodcapturesinteractionsthatthestandardRFmethoddoesnot.
Categories: Bioinformatics,Diabetes,Gene‐EnvironmentInteraction,MachineLearningTools,
PsychiatricDiseases
P114
StudyoFHumanMGPpromotervariantsinCADpatients:From
Experimenttoprediction
BitaSadatHosseini1,AbazarRoustazadeh2,MohammadNajafi3
1BiochemistryDepartment,IranUniversityofMedicalSciences,Tehran,Iran
2JahromUniversityofMedicalSciences,Jahrom,Iran
3BiochemistryDepartment,CellularandMolecularResearchCenter,IranUniversityofMedicalSciences,
Tehran,Iran
Background:MatrixGlaprotein(MGP)isknownasacalciumscavengerwithinsub‐endothelial
spaceofvesselsandissuggestedtoreducetheriskofcoronaryarterydiseases.Inthisstudy,we
comparedtheMGPpromoterhighminorallelefrequency(MAF)variantsandthechangesonthe
predictedtranscriptionfactorelementsinpatientswithcoronaryarterydisease.Methods:TheMGP
promotergenotypesandhaplotypesweredetectedbyARMS‐RFLPPCRtechniques.TheJaspar
profiles(similarity>80)wereusedforscoringthepolymorphicvariantswithinthetranscription
factorelements.Results:TheMGPpolymorphichaplotypesandgenotypeshadnotsignificant
differencesbetweencontrolandpatientgroups(P=0.4andP=0.1respectively).Furthermore,the
resultsshowedthatthegenotypeandhaplotypedistributionsoftheMGPpromoterhigh‐MAF
polymorphisms,asconfirmedinthepredictionstudiesarenotsignificantlyassociatedwiththe
coronaryarterydisease.Discussion:Thepredictionandpopulationresultsshowedthattheallele
changeswithintheelementshavenotsignificantlyrelatedtothetranscriptionfactorscoresand
stenosisofcoronaryarteries.
Categories: Bioinformatics,CardiovascularDiseaseandHypertension,HaplotypeAnalysis
P115
Anovelfunctionaldataanalysisapproachtodetectinggeneby
longitudinalenvironmentalexposureinteraction
PengWei1
1UniversityofTexasSchoolofPublicHealth
Mostcomplexdiseasesarelikelytheconsequenceofthejointactionsofgeneticandenvironmental
factors.Identificationofgene‐environment(GxE)interactionsnotonlycontributestoabetter
understandingofthediseasemechanisms,butalsoimprovesdiseaseriskpredictionandtargeted
intervention.Incontrasttothelargenumberofgeneticsusceptibilitylocidiscoveredbygenome‐
wideassociationstudies,therehavebeenveryfewsuccessesinidentifyingGxEinteractionswhich
maybepartlyduetolimitedstatisticalpowerandinaccuratelymeasuredexposures.Whileexisting
statisticalmethodsonlyconsiderinteractionsbetweengenesandstaticenvironmentalexposures,
manyenvironmentalfactors,suchasairpollutionanddiet,changeovertime,andcannotbe
accuratelycapturedatonemeasurementtimepoint.Thereisadearthofstatisticalmethodsfor
detectinggenebytime‐varyingenvironmentalexposureinteractions.Hereweproposeapowerful
functionallogisticregression(FLR)approachtomodelthetime‐varyingeffectoflongitudinal
environmentalexposureanditsinteractionwithgeneticfactorsondiseaserisk.Capitalizingonthe
powerfulfunctionaldataanalysisframework,ourproposedFLRmodeliscapableofaccommodating
longitudinalexposuresmeasuredatirregulartimepointsandcontaminatedbymeasurement
errors.WeusesimulationstoshowthattheproposedmethodcancontroltheTypeIerrorandis
morepowerfulthanalternativeadhocmethods.Wedemonstratetheutilityofthisnewmethod
usingdatafromacase‐controlstudyofpancreaticcancertoidentifythewindowsofvulnerabilityof
lifetimebodymassindexontheriskofpancreaticcanceraswellasgeneswhichmaymodifythis
association.
Categories: Cancer,Gene‐EnvironmentInteraction
P116
LeveragingFamilyStructurefortheAnalysisofRareVariantsinKnown
CancerGenesfromWESofAfricanAmericanHereditaryProstateCancer
CherylDCropp1,ShannonKMcDonnell2,SumitMiddha2,DanielleKaryadi3,StephenNThibodeau4,
JanetStanford5,KathleenACooney6,JoanEBailey‐Wilson1,JohnDCarpten7,fortheInternational
ConsortiumofProstateCancerGenetics
1ComputationalandStatisticalGenomicsResearchBranch,NationalHumanGenomeResearch
Institute/NationalInstitutesofHealth,Baltimore,MD
2DepartmantofHealthScienceResearch,MayoClinic,Rochester,MN
3CancerGeneticsBranch,NationalHumanGenomeResearchInstitute/NationalInstitutesofHealth,Bethesda,
MD
4DepartmantofLaboratoryMedicineandPathology,MayoClinic,Rochester,MN
5PublicHealthSciencesDivision,EpidemiologyProgram,FredHutchinsonCancerResearchCenter,Seattle,
WA
6UniversityofMichiganComprehensiveCancerCenter,AnnArbor,MI
7IntegratedCancerGenomicsDivision,TranslationalGenomicsResearchInstitute(TGen),Phoenix,AZ
“LeveragingFamilyStructurefortheAnalysisofRareVariantsinKnownCancerGenesfromWESof
AfricanAmericanHereditaryProstateCancer”Prostatecancer(PRCA)isthesecondleadingcauseof
cancerdeathinNorthAmericanmenanditdisproportionatelyaffectsAfricanAmerican(AA)men,
whohavehigherincidenceandmortalityratescomparedtomenwithoutknownAfricanancestry.
DisentanglingtheenvironmentalandgeneticfactorsinAAwithhereditaryPRCAremainselusive.
TheAfricanAmericanHereditaryProstateCancerStudy(AAHPC)wasdevelopedtofurtherexplore
theroleofgeneticsinthecausationofhereditaryPRCAinAA.AAHPCisinpartnershipwiththe
InternationalConsortiumforProstateCancerGenetics(ICPCG)toconductcollaborativestudiesin
PRCAgeneticsinmultiplexfamilies.AspartofanICPCGsequencingstudyof539affected
individualsfrom366PRCApedigrees,weperformedwholeexomesequencingon16AAHPC
affectedmenfrom12pedigrees.Post‐variantcallingqualitycontrolwasimplementedusingGolden
HelixSVS8softwarewithfilterssetforremovalofvariantswithReadDepth<10,QualityScore<
20,QualityScore:ReadDepthRatio<0.5,CallRate<0.75.Variantswereadditionallyfilteredby
minorallelefrequency(MAF)basedontheNHLBIESP650051‐V2exomesvariantfrequenciesfor
AApopulationusingaMAFthresholdof1%.AfterQC,174,047variantsremainedforfurther
analysis.Intheseanalyses,wefocusedon13knowncancercausinggenes.TwoAAHPCfamilieshad
>1affectedmemberssequenced(3perfamily).Underadominantmodel,Family1shared14
variantsinthesegenesamongallaffectedswhileFamily2shared17variantsamongallaffected
men.Additionalstudiesareunderwaytodetermineifpredicteddamagingvariantsinthesegenes
aresharedinotherICPCGAAfamiliestohelpunravelthegeneticheterogeneityofhereditaryPRCA
inAA.
Categories: Cancer,DataMining,GenomicVariation,SequencingData
P117
Associationofbreastcancerrisklociwithsurvivalofbreastcancer
patients
MyrtoBarrdahl1,FedericoCanzian2,SaraLindström3,IreneShui3,RudolfKaaks1,DanieleCampa1
1DivisionofCancerEpidemiology,GermanCancerResearchCenter(DKFZ),Heidelberg,Germany
2GenomicEpidemiologyGroup.GermanCancerResearchCenter(DKFZ),Heidelberg,Germany
3DepartmentofEpidemiology,HarvardSchoolofPublicHealth,BostonMA,USA
Thesurvivalofbreastcancerpatientsislargelyinfluencedbytumorcharacteristics,suchasTNM
stage,tumorgradeandhormonereceptorstatus.However,thereisgrowingevidencethatinherited
geneticvariationmightinfluencethediseaseprognosisandresponsetotreatment.Severallinesof
evidencesuggestthatpolymorphismsinfluencingbreastcancerriskmightalsobeassociatedwith
breastcancersurvival.Withtheaimoffurtherexploringthispossibility,weselected35
polymorphismsassociatedwithbreastcancerriskandinvestigatedtheirroleinthediseaseover‐all
survival.Westudied10,255breastcancerpatientsfromtheNationalCancerInstituteBreastand
ProstateCancerCohortConsortium(BPC3)ofwhich1,379hadfatalbreastcancer.Wealso
conductedameta‐analysisofalmost35,000patientsand5,000deaths,combiningresultsfromthe
currentstudyandfromtheBreastCancerAssociationConsortium(BCAC).InBPC3weobserveda
significantassociationbetweentheCalleleofLSP1‐rs3817198andreduceddeathhazard(HRper‐
allele=0.70;95%CI:0.58‐0.85;Ptrend=2.84×10‐4).Thisassociationwassupportedbythe
observationthattheCalleleofthisSNPincreasestheexpressionofthetumorsuppressorcyclin‐
dependentkinaseinhibitor1C(CDKN1C).Themeta‐analysisshowedasignificantassociation
betweenTNRC9‐rs3803662andanincreaseddeathhazard(HRMETA=1.21;95%CI:1.09‐1.35;
P=2.47×10‐4comparinghomozygotesfortheminorallelevs.homozygotesforthemajorallele).In
conclusion,weshowthatthereislittleoverlapbetweenSNPsassociatedwithbreastcancerriskand
SNPsassociatedwithbreastcancerprognosis,withthepossibleexceptionsofLSP1‐rs3817198and
TNRC9‐rs3803662.
Categories: Cancer
P118
Evidenceofgene‐environmentinteractionsinrelationtobreastcancer
risk,resultsfromtheBreastCancerAssociationConsortium
MyrtoBarrdahl1,AnjaRudolph1,NickOrr2,PaulPharoah3,PerHall4,MontserratGarcia‐Closas5,
MarjankaSchmidt6,RogerMilne7,DougEaston8,JennyChang‐Claude1
1DivisionofCancerEpidemiology,GermanCancerResearchCenter(DKFZ),Heidelberg,Germany
2BreakthroughBreastCancerResearchCentre,InstituteofCancerResearch,London,UK
3CentreforCancerGeneticEpidemiology,DepartmentofPublicHealthandPrimaryCare,Universityof
Cambridge,Cambridge,UK
4MedicalEpidemiologyandBiostatistics,KarolinskaInstitutet,Stockholm,Sweden
5SectionsofEpidemiologyandGenetics,InstituteofCancerResearchandBreakthroughBreastCancer
ResearchCentre,London,UK
6DivisionofMolecularPathologyandDivisionofPsychosocialResearchandEpidemiology,Netherlands
CancerInstitute,Amsterdam,TheNetherlands
7GeneticandMolecularEpidemiologyGroup,HumanCancerGeneticsProgramme,SpanishNationalCancer
ResearchCentre(CNIO),Madrid,Spain
8DivisionofCancerEpidemiologyandGenetics,NationalCancerInstitute,NIH,Bethesda,Maryland,USA
Severalnewsusceptibilityallelesforbreastcancer(BC)riskhavebeenidentifiedbytheBreast
CancerAssociationConsortium(BCAC)throughimputationofgeneticdatato1000Genomesand
fine‐mappingofknownsusceptibilityloci.Weinvestigatedwhethertheidentifiedsinglenucleotide
polymorphism(SNP)associationsaremodifiedbyestablishedBCriskfactors.Weassessed
multiplicativeinteractionbetween12BCriskfactorsand74SNPs,ofwhich54wereimputed(from
17knownregions)and29genotyped(in22regions).Weuseddatafromupto25,539invasiveBC
casesand29,664controlsfrom21studiesinBCAC.Theriskfactorswere:ageatmenarche,parity,
numberoffull‐termpregnancies(FTP),ageatfirstFTP,breastfeeding,BMI,height,oral
contraceptiveuse,currentpostmenopausalhormoneuse(estrogenandestrogen‐progesterone),
currentsmokingandcumulativelifetimealcoholintake.InteractionsbetweenSNPsandBCrisk
factorswereevaluatedusinglikelihood‐ratioteststocomparelogisticregressionmodelswithand
withoutinteractionterms.Allmodelswereadjustedforstudy,ageandancestryinformative
principalcomponents.WefoundasuggestiveinteractionbetweenaSNPinthe9q31regionand
currentsmoking(Pinteract=5.3×10‐5),whichwassignificantafterBonferronicorrectionofthe
significancethreshold(P<5.6×10‐5).Inparticular,theG‐allelewasinverselyassociatedwithBCrisk
amongsmokers(ORper‐allele:0.68,95%CI:0.57‐0.81,P=1.7×10‐5)butnotamongnon‐smokers
(ORper‐allele:0.95,95%CI:0.89‐1.03,P=0.2).Inconclusion,thefindingsofourstudyprovide
indicationsthattheassociationbetweencommongeneticvariantsandBCriskmayvaryacrossthe
levelsoftheBCriskfactors.
Categories: Cancer,Gene‐EnvironmentInteraction
P119
Integrationofpathwayandgene‐geneinteractionanalysesreveal
biologicallyrelevantgenesforBreslowthickness,amajorpredictorof
melanomaprognosis
AmauryVaysse1,ShenyingFang2,MyriamBrossard1,WeiVChen2,HamidaMohamdi1,EveMaubec1,
Marie‐FrançoiseAvril3,ChristopherIAmos4,JeffreyELee5,FlorenceDemenais1
1INSERMU946,Paris,France;UniversitéParisDiderot,Paris,France;
2MDAndersonCancerCenter(MDACC),Houston,Texas,USA;
3HôpitalCochin,UniversitéParisDescartes,Paris,France
4GeiselCollegeofMedicine,DartmouthCollege,NewHampshire,USA
5MDAndersonCancerCenter(MDACC),Houston,Texas,USA
Breslowthickness(BT),ameasureofinvasionofmelanomaintheskin,isamajorpredictorof
melanomasurvival.Todate,thegeneticfactorsunderlyingBTarelargelyunknown.Weconducteda
GWASofBTintheFrenchMELARISKstudy(966cases)andtheUSMDACCstudy(1546cases).We
firstperformedsingle‐SNPanalysisthatwasfollowedbymulti‐markeranalysistocharacterize
pathwaysandgene‐geneinteractionsassociatedwithBT.Pathwayanalysiswasbasedonthegene
setenrichmentanalysis(GSEA)method,usingtheGeneOntology(GO)database.Allgenepairs
withineachofthemelanoma‐associatedGOsweretestedforinteractionusingalinearregression
model.SingleSNPanalysisofHapmap3‐imputedSNPsinMELARISKshowedevidenceforfiveloci
thatreachedP<10‐5butnoneoftheseassociationswasreplicatedinMDACC,suggestingthe
existenceofmanyvariantswithsmalleffect.IntheGSEAanalysis,onemillionimputedSNPswere
assignedto22,000genes,whichwereassignedto316Level4‐GOcategories.ThreeGOcategories
werefoundtobeenrichedingenesassociatedwithBT(FDR≤0.05inbothstudies):hormone
activity,cytokineactivityandmyeloidcelldifferentiation.Atotalof61genesweredrivingthese
pathways.Interestingly,expressionoffourofthesegenes(CXCL12,TNFSF10,VEGFA,CDC42)was
reportedtobeassociatedwithmelanomaprogressionintumors.Cross‐geneSNP‐SNPinteraction
analysiswithineachofthethreeidentifiedGOsshowedevidenceforinteractionforthreeSNPpairs
(P≤10‐4inMELARISKandreplicationatP≤0.05inMDACC).Oneofthesegenepairs(SCINxCDC42,
combinedP=2x10‐6)hasbiologicalrelevancesinceSCINandCDC42proteinsareinvolvedinthe
actindynamicswithoppositeroles.Funding:INCa_5982
Categories: Cancer,Gene‐GeneInteraction,MultilocusAnalysis,Pathways,QuantitativeTraitAnalysis
P120
JAG1polymorphismisassociatedwithincidentneoplasminasouthern
Chinesepopulation
Chor‐WingSing1,VivianWai‐YanLui1,Pak‐ChungSham1,KathrynChoon‐BengTan1,AnnieWai‐
CheeKung1,IanChi‐KeiWong1,BernardMan‐YungCheung1,JohnnyChun‐YinChan1,Ching‐Lung
Cheung1
1TheUniversityofHongKong
Aim:Jagged1(JAG1)isaligandofnotchreceptorsthatregulatescelldivision,differentiation,and
survival.Over‐expressionofJAG1hasbeenlinkedtoincreasedriskofcancer.Wepreviouslyshowed
thatrs2273061ofJAG1wasassociatedwithbonemineraldensity(BMD)usinggenome‐wide
association,andtheSNPwasassociatedwithJAG1expressioninboneandbloodcells.We
hypothesizedthatthisSNPhasassociationwithneoplasm.Methods:TheSNPrs2273061ofJAG1
wasgenotypedintwoindependentcohortswithouthistoryofneoplasmatbaseline.Thecohorts
werefollowed(median10.8years)fordevelopmentofneoplasmusingelectronicmedicaldatabase
oftheHongKongHospitalAuthority.AscertainmentofneoplasmwasbasedonICD9code140‐239.
Coxproportionalhazardsregressionmodelsadjustedforage,sex,BMI,andlumbarspineBMDZ‐
scorewereusedforassociationanalysis.AUCwasusedtotestpredictiveaccuracyofthemodels.
Result:Inthefirstcohort(n=731;80incidents;7620person‐year),minorallele(G)ofrs2273061
wassignificantlyassociatedwithneoplasm(HR=0.68;95%CI:0.47‐0.98).Theresultwasvalidated
(HR=0.81;95%CI:0.66‐0.99)inreplicationcohort(n=1,885;241incidents;19966person‐year).
Meta‐analysisshowedamoresignificantassociation(HR=0.78;95%CI:0.65‐0.93;p=0.005).AUCfor
basicclinicalmodel(age+sex+BMI)inpredictingneoplasmwas0.618(95%CI:0.588‐0.648).The
additionofrs2273061genotypetothebasicmodelincreasedAUCto0.635(95%CI:0.605‐0.665),
andtheincrementwasstatisticallysignificant.Conclusion:JAG1polymorphismhasassociationwith
incidentneoplasmHowever,furtherstudyisrequiredtoevaluateanyfunctionaleffectsof
rs2273061ontumorformation.
Categories: Cancer
P121
Epigenome‐widemethylationarrayanalysisrevealsfewmethylation
patterndifferencesbetweenhyperplasticpolypsandsessileserrated
adenomas/polyps
JingLi1,AngelineSAndrew1,AmitabhSrivastava2,JasonHMoore1
1InstituteforQuantitativeBiomedicalSciences,DartmouthCollege
2BrighamandWomen'sHospital
Thecolorectal‘serratedpolyps’ariseviaaneoplasticpathwayandwerehistoricallynotconsidered
withmalignantpotential.Majorsubtypesofserratedcolorectalpolyps,hyperplasticpolyps(HPs)
andsessileserratedadenomas/polyps(SSA/Ps),areclassifiedbasedonmorphologicaldistinctions.
RecentstudieshaveidentifiedSSA/Pasahigh‐risksubtypeofserratedcolorectalpolypsthatcan
developintocolorectalcancer.OurgoalwastodeterminewhetherHPsandSSA/Pshavedistinct
underlyingDNAmethylationsignatures.Toevaluatethesubtype‐specificDNAmethylationstatus,
DNAsfrom35HPsand42SSA/PswereextractedandIlluminaInfiniumHumanMethylation450
BeadChiparrayswereusedtoprofilethemethylationstatusfor>485,000CpGloci.Principal
componentanalysisrevealedthatthetopprincipalcomponents,whichaccountforthelargest
amountofvariabilityofmethylationstatus,arenotsignificantlyassociatedwithsubtype(p=0.414).
Also,linearmixed‐effectsmodelsshowedthatthemethylationpatternisnotsignificantlydifferent
betweensubtypes,aftercontrollingforage,gender,polypsize,anatomicsideandbatch.Wealso
comparedSSA/PsandHPsusingtheprobesthatmaptotheCIMPpanellociandfoundno
statisticallysignificantdifferencesinmethylationstatusbymorphology.Comparingthenormalvs.
serratedcolorectalpolypsrevealed18probeswithsignificantlydifferentmethylationlevelsbelow
theBonferronithreshold(p=<1.06e‐7).Ourresultssuggeststhatdysregulatedmethylationis
prevalent,involvinganumberofnon‐CIMPCpGsandlikelyoccursearlyinserratedneoplasia.The
datadonotsupportthehypothesisthatSSA/PsandHPsariseviadifferentepigeneticpathways.
Categories: Cancer,EpigeneticData,Epigenetics
P122
Theeffectofbileacidsequestrantsontheriskofcardiovascularevents:
Ameta‐analysisandMendelianRandomizationanalysis
GuillaumePare1,StephanieRoss1,MatthewD'Mello1,SoniaSAnand1,JohnEikelboom1,AlexanderFR
Stewart2,NileshJSamani3,RobertRoberts2
1McMasterUniversity
2UniversityofOttawa
3UniversityofLeicester
Statinsareusedtolowerlowdensitylipoproteincholesterol(LDL‐C)buttheymaybepoorly
toleratedorineffective.Bileacidsequestrants(BAS)acttoreducetheintestinalabsorptionof
cholesterolbutprevioustrialswereunderpoweredtodemonstrateaneffectonclinicaloutcomes.
Weconductedasystematicreviewandmeta‐analysisofrandomizedcontrolledtrials(RCTs)to
assesstheeffectoftwoapprovedBAS,cholestyramineandcolesevelam,onplasmalipidlevels.We
thenappliedtheprinciplesofMendelianRandomizationtoestimatetheeffectofBASonreducing
theriskofcoronaryarterydisease(CAD)byquantifyingtheeffectofrs4299376(ABCG5/ABCG8),
whichaffectstheintestinalcholesterolabsorptionpathwaytargetedbyBAS,onbothLDL‐Cand
CAD.NineteenRCTswithatotalof7,021studyparticipantsmettheinclusioncriteria.
Cholestyramine24g/dwasassociatedwitha23.5mg/dLreductioninLDL‐C(95%CI:‐26.8,‐20.2;
N=3,806)andatrendtowardsreducedriskofCAD(OR:0.81,95%CI:0.70‐1.02;P=0.07;N=3,806)
whilecolesevelam3.75g/dwasassociatedwitha22.7mg/dLreductioninLDL‐C(95%CI:‐28.3,‐
17.2;N=759).Basedonthegeneticassociationofrs4299376witha2.75mg/dLdecreaseinLDL‐C
anda5%decreaseinriskofCADoutcomes,weestimatedthatcholestyraminemaybeassociated
withanORforCADof0.63(95%CI:0.52‐0.77;P=6.3x10‐6;N=123,223)andcolesevelamwithan
ORof0.64(95%CI:0.52‐0.79,P:4.3x10‐5).Theseestimateswerenotstatisticallydifferentfrom
previouslyreportedtrendsfromBASclinicaltrials(P>0.05).ThecholesterolloweringeffectofBAS
canthusbeexpectedtotranslateintoaclinicallyrelevantreductionintheriskofCAD.
Categories: CardiovascularDiseaseandHypertension,MendelianRandomisation
P123
MendelianRandomisationstudyofthecausalinfluenceofkidney
functiononcoronaryheartdisease
PimphenCharoen1,UCLEBConsortium,Juan‐PabloCasas1,DorotheaNitsch1,FrankDudbridge1
1DeptNon‐communicableDiseaseEpidemiology,LondonSchoolofHygieneandTropicalMedicine,London,
UK
Kidneyfunctionisknowntocorrelatewithcoronaryheartdisease(CHD).Howeveritisnotyetclear
whetherkidneyfunctionreflectsacausalpathwaybecausethisobservedassociationcoulddueto
otherconfoundingfactors,suchasBMIandbloodpressure.ThereforeweappliedMendelian
Randomisation(MR)whichallowsdisentanglingofcauseandeffectinthepresenceofpotential
confounding,todeterminewhetherkidneyfunctionhascausalroletoCHD.Toourknowledge,this
isthefirstMRstudytoinvestigatethecausalinfluenceofkidneyfunctiononCHD.Thelevelof
kidneyfunctionwasmeasuredbyanestimatedglomerularfiltrationrate(eGFR).Toenhancethe
statisticalpowerbyincreasingthesamplesizeupto200K,thesummarystatisticsofassociations
betweengeneticvariantsandCHDfromourUCL‐LSHTM‐Edinburgh‐Bristol(UCLEB)Consortium
werecombinedwiththeCARDIoGRAMplusC4DConsortiumwhichisavailablepublicly.Eighteen
SNPspreviouslyreportedtobeassociatedwitheGFRwerethenestablishedasinstrumentswhere
theircausaleffectscanbecombinedusingthemethodproposedbyBurgessetal.2013aswellasa
moregeneralmodelwhichallowsflexiblescalingonanestimatedcausaleffect.Weobservedno
significantevidenceofcausalinfluenceofeGFRonCHD.Thismaybeduetothelimitedexplanatory
powerofourgeneticinstrument,despiteourlargesamplesize,butalsoimpliesthattheassociation
observedbetweenkidneyfunctionandCHDcouldduetoconfoundingfactorsorreversecausation.
Categories: CardiovascularDiseaseandHypertension,Causation,MendelianRandomisation
P124
Sharedgeneticriskofmyocardialinfarctionandbloodlipidsusing
empiricallyderivedextendedpedigrees:resultsfromtheBusselton
HealthStudy
GemmaCadby1,PhillipEMelton1,JennieHui2,JohnBeilby3,ArthurWMusk4,AlanLJames5,Joseph
Hung6,JohnBlangero7,EricKMoses1
1CentreforGeneticOriginsofHealthandDisease,UniversityofWesternAustralia
2BusseltonPopulationMedicalResearchInstituteInc
3PathWestLaboratoryMedicineWA
4DepartmentofRespiratoryMedicine,SirCharlesGairdnerHospital
5DepartmentofPulmonaryPhysiologyandSleepMedicine,SirCharlesGairdnerHospital
6SchoolofMedicineandPharmacology,UniversityofWesternAustralia
7TexasBiomedicalResearchInstitute
Quantitativeendophenotypesrelatedtocomplexdiseasesprovideincreasedpowerforgene
localisationandidentificationcomparedwithdichotomousdiseasestatus.Inthisstudy,we
employedempiricallyderivedidenticalbydescent(IBD)measurestoestimatetheheritabilitiesand
geneticcorrelationsbetweenbloodlipidendophenotypes(HDL‐C,LDL‐Candtriglycerides)and
myocardialinfarction(MI)in4671individualswhoattendedthe1994/95BusseltonHealthStudy
(BHS).IBDestimateswerederivedfromgenome‐wideassociationdatausingLDAKsoftware.MI
eventswereidentifiedfromhospitalmorbidityanddeathregistrydataobtainedfromtheWestern
AustralianHealthDepartmentDataLinkageUnit.Heritabilityandgeneticcorrelationbetweentraits
werecalculatedafteradjustingforsignificantcovariates(e.g.age,sex,lipidmedication,smoking
status).Approximately75%ofthe4671individualswererelatedtoatleastoneotherBHS
participant(uptoandincludingthirddegreerelatives).Between1970and2011,331individuals
hadatleastoneMIevent.HeritabilityofHDL‐C,LDL‐Candtriglycerideswere0.54,0.48,and0.34,
respectively(allP<0.001).HDL‐Candtriglyceridesbothshowedasignificantsharedgenetic
correlationwithMIof‐0.43(P=0.01)and0.46(P=0.03),respectively.HDL‐C,LDL‐Cand
triglycerideswerehighlyheritableintheBHSandsimilartoearlierreportedestimates,
demonstratingtheviabilityofusingempiricallyderivedIBDs.HDL‐Candtriglyceridesbothshowed
geneticcorrelationwithMI,suggestingthesearevaluableendophenotypesforCVD‐riskgene
discovery.
Categories: CardiovascularDiseaseandHypertension,Heritability
P125
AnalysisofCase‐Base‐Controldesigns
NajlaSElhezzani1,WicherPBergsma2,MikeWeal3
1King'scollegeLondonandKingSauduniversity
2TheLondonschoolofeconomics
3King'scollegeLondon
Case‐controlstudiescompareindividualswithatraitofinterest(cases)withotherswhodon'thave
it(controls).However,Inmanygeneticsassociationstudiesthecontrolgroupistakenasasample
fromthepopulationwhereindividualshaveunknowntraitstatus(bases).Thisapproachappeared
tobesuccessfulwhenthetraitisrare.However,iftheprevalenceishighthenusingthebasesasa
setofcontrolswillleadtounreliableresultsaspoweriscompromisedinthiscase.Accordingly,we
proposedthecase‐base‐controldesignwhichallowsthethreesampletypestobeusedinasingle
analysis.Totestwethergenotypefrequenciesdifferbetweencasesandcontrolstakingintoaccount
thebases,wederivedthescoretest.ThetestreducestoCochran‐ArmitagetestwhentheCBC
reducestotheCC.Thescorestatisticsshowsagoodadherencetotheasymptoticdistribution.We
investigatedthemaximumlikelihoodestimatesoftheunderlyingparametersanalyticallyand
numericallyusingexpectation‐maximizationalgorithm.WederivedtheWald’sandlikelihoodratio
tests.Wefoundthatusingmoderatesamplesizes,LRTwasslightlymorepowerfulthanothers,
Howeverforlargesamplesthepowerofalltestsnotonlybecomessimilarbutalsoindependentof
theprevalence.Finally,wecomparedtheCBCdesignwiththeusualcase‐controldesign.Wefound
thatonlyiftheprevalenceiswell‐specifiedandtheproportionofcasesinthebasesisdifferentfrom
thatintheexperiment(casesandcontrols),thentheCBCwouldprovidemorepowercomparedto
theCC.Lookingatthecaseofhavingalargesetofbases,wefoundthatifprevalenceiswellspecified
then,theoptimaldesignwillbegainedbyusingonlycasesifprevalenceislowandonlycontrolsifit
ishigh.
Categories: Case‐ControlStudies,MaximumLikelihoodMethods,PopulationGenetics,Prediction
Modelling,SampleSizeandPower
P126
PolymorphismsinHTR3A,CYP1A2,DRD4andCOMTandresponseto
clozapineintreatment‐resistantschizophrenia:agene‐geneinteraction
analysis
RVeeraManikandan1,AntoPRajkumar2,LakshmikirupaSundaresan1,ChithraC1,AnjuKuruvilla1,
AlokSrivastava1,PoonkuzhaliBalasubramanian1,KuruthukulangaraSJacob1,MollyJacob1
1ChristianMedicalCollege,Vellore,India
2AarhusUniversity,Aarhus
Variableresponsestoclozapineinpatientswithschizophreniaarecomplexandpoorlyunderstood
phenomena.Thefindingsofpharmacogeneticstudiesontheuseofthisdrugarepoorlyreplicated.
Effectsofindividualpolymorphismshaverarelyprovedexplanatory.Onepossibleexplanationmay
bemulti‐factorialinvolvementofgeneticandenvironmentalinfluences.Theaimofthisstudyisto
evaluatetheroleofpossiblesecondandthirdordergeneticinteractions(epistasis)between
polymorphismsinCYP1A2(*1F,*1D,*1E,*1C),HTR3A(rs1062613andrs2276302),DRD4(120‐bp
duplication)andCOMT(Val158Met)genesoverclinicalresponse,serumlevelsandadverseeffects
ofclozapineinpatientswithtreatment‐resistantschizophrenia(TRS).Themodel‐based
multidimensionalityreduction(MB‐MDR)methodhasrecentlybeenshowntobesuperiorto
traditionalparametricregressionmethodsindetectinghigherordergene‐geneinteractions.We
usedthisapproachinasampleof93patientswithTRStoexploretheepistaticeffectsofthe
polymorphismsofinterestonclinicalphenotypesofclozapine.TheMB‐MDRanalysisshoweda
significantinteractionbetweenVal158Met,CYP1A2*1Dandrs1062613polymorphismsandclinical
responsetoclozapine(p=0.002).Inaddition,multiplesignificantsecondandthirdorder
interactionswereobservedwithregardtotheadverseeffectsofclozapine(p<0.05).Allthereported
interactionswerefoundtobesignificantafter1000permutations.Theobservedmultiplesignificant
interactionsemphasizestheimportanceofepistaticanalysisinpharmacogeneticstudiesof
clozapine.Suchanapproachmaybeusefulinpredictingapatient’sresponsetoclozapinetherapy.
Categories: Case‐ControlStudies,Gene‐GeneInteraction,GenomicVariation,MultifactorialDiseases,
MultilocusAnalysis,MultivariatePhenotypes,PopulationGenetics,PsychiatricDiseases,Quantitative
TraitAnalysis
P127
Jointmodelingoflongitudinalandtime‐to‐eventphenotypesingenetic
associationstudies:strengthsandlimitations
OsvaldoEspin‐Garcia1,2,ZhijianChen2,AndrewDPaterson3,ShelleyBBull1,2
1DallaLanaSchoolofPublicHealth,UniversityofToronto
2Lunenfeld‐TanenbaumResearchInstituteofMountSinaiHospital
3TheHospitalforSickChildren;DallaLanaSchoolofPublicHealth,UniversityofToronto
Genome‐wideassociationstudydesignsthatevaluatemultipleendpointsinobservationalsettings
arebecomingmorecommon.Whileoftentimesexaminationofsingleoutcomesissufficientforthe
purposesofthestudy,therearecaseswherejointanalysisisinformativeinthesimultaneous
evaluationofgeneticassociationwithmultipleendpoints.Inparticular,thestudyoftime‐to‐event
andlongitudinaldataarisesnaturallyincohortstudies,buttheuseofjointanalysishasremained
ratherunexploredingeneticassociation.Themotivationforjointanalysiscomestolightunder
differentscenarios.TheobjectivemaybetodistinguishwhetheraSNPhasadirectassociationwith
atime‐to‐eventphenotype,and/oranindirectassociationthroughanintermediatequantitative
trait(QT).Thiscanbethoughtofasaformofcausalinference:iftheSNPassociationwithtime‐to‐
eventisnegligiblewhentheQTiswellmodelledinthesurvivalanalysis,thenitcannothaveadirect
causaleffectontimetoevent.Alternatively,geneticassociationwithaQTmaybeofprimary
interest,butaclinicaleventcausesinformativecensoringofthetrait.Inthiswork,wefocusonthe
jointmodelproposedbyWulfsohnandTsiatis(1997)andwidelyusedinclinicalstudiesofCD4+
countsandtimetoAIDS.Wediscussestimationandcausalinterpretationofgeneticassociation
parametersinthejointmodel,examinestatisticalpropertiessuchasefficiencyandbiasoftheeffect
estimatescomparedtotheirsingle‐outcome‐analysiscounterpart,andquantifypotential
improvementinpowertodetectgeneticassociation.Inaddition,wereviewsoftware
implementationandcomputationalfeasibilityinthecontextofgenome‐wideanalysis.
Categories: Causation,MaximumLikelihoodMethods,MultivariatePhenotypes,QuantitativeTrait
Analysis
P128
Perinataldepressionandomega‐3fattyacids:AMendelian
randomisationstudy
HannahSallis1,2,ColinSteer3,LaviniaPaternoster1,GeorgeDaveySmith1,JonathanEvans2
1MRCIntegrativeEpidemiologyUnit,SchoolofSocialandCommunityMedicine,UniversityofBristol,UK
2CentreforAcademicMentalHealth,SchoolofSocialandCommunityMedicineUniversityofBristol,UK
3CentreforChildandAdolescentHealth,SchoolofSocialandCommunityMedicine,UniversityofBristol,UK
IntroductionTherehavebeennumerousstudiesinvestigatingtheassociationbetweenomega‐3
fattyacids(FAs)anddepression,withmixedfindings.Weproposeanapproachwhichislargelyfree
fromissuessuchasconfoundingorreversecausalitytoinvestigatethisrelationshipusing
observationaldatafromapregnancycohort.MethodsTheAvonLongitudinalStudyofParentsand
Children(ALSPAC)cohortcollectedinformationonFAlevelsfromantenatalbloodsamplesand
depressivesymptomsatseveraltimepointsduringpregnancyandthepostnatalperiod.
ConventionalepidemiologicalanalyseswereusedinadditiontoaMendelianrandomisation(MR)
approachtoinvestigatetheassociationbetweenlevelsoftwoomega‐3FAs(docosahexaenoicacid
(DHA)andeicosapentaenoicacid(EPA))andperinatalonsetdepression,antenataldepressionand
postnataldepression.WeconstructedaweightedalleleriskscoreusingindependentSNPs
identifiedasassociated(p<5x10‐6)inarecentgenome‐wideassociationstudyofomega‐3FAsby
theCHARGEconsortium.ResultsandDiscussionWeakevidenceofapositiveassociationwithboth
EPA(n=2377;OR=1.07;95%CI:0.99‐1.15)andDHA(n=2378;OR=1.08;95%CI:0.98‐1.19)with
perinatalonsetdepressionwasfoundusingamultivariablelogisticregressionadjustingforsocial
classandmaternalage.However,thestrengthofassociationwasfoundtoattenuatewhenusingan
MRanalysistoinvestigateDHA.Inconclusion,wefoundweakevidenceofapositiveassociation
betweenomega‐3FAsandperinatalonsetdepression.However,withoutconfirmationfromtheMR
analysis,weareunabletodrawconclusionsregardingcausality.
Categories: Causation,MendelianRandomisation,PsychiatricDiseases
P131
AGene‐EnvironmentInteractionBetweenCopyNumberBurdenand
OzoneExposureProvidesaHighRiskofAutism
DokyoonKim1,HeatherVolk2,SarahPendergrass1,MollyAHall1,ShefaliSVerma1,Santhosh
Girirajan1,IrvaHertz‐Picciotto3,MarylynRitchie1*,ScottSelleck1
1DepartmentofBiochemistry&MolecularBiology,thePennsylvaniaStateUniversity,UniversityPark,PA
2DepartmentofPreventiveMedicine,KeckSchoolofMedicine,UniversityofSouthernCalifornia,LosAngeles,
CA;DepartmentofPediatrics,Children’sHospitalLosAngeles,UniversityofSouthernCalifornia,LosAngeles,CA
3DepartmentofPublicHealthSciences,UniversityofCalifornia,Davis,Davis,CA
Autismisadisorderofneuraldevelopmentasacomplexgenetictraitwithahighdegreeof
heritabilityaswellasadocumentedsusceptibilityfromenvironmentalfactors.Therelative
contributionsofgeneticfactors,environmentalfactorsandtheinteractionsbetweenthemtoarisk
ofautismarepoorlyunderstood.Whilemostautismrelatedcopynumbervariations(CNV)
identifiedtodate,eachwithasubstantialrisk,arehighlypenetrantforthisdisorder,theyconstitute
rareeventscontributingmodestlytotheoverallheritability.Genome‐wideanalysisofCNVhas
demonstratedacontinuousriskofautismassociatedwiththelevelofcopynumberburden,
measuredastotalbasepairsofduplicationordeletion.Inaddition,environmentalexposuretoair
pollutantshasbeenidentifiedasariskfactorfordevelopingautism,includingparticulatepollutants
andnitrogendioxide.WehaveexaminedtherelativecontributionofCNV(measuredastotalbase
pairsofcopynumberburden),exposuretoairpollution,andtheinteractionbetweenairpollutant
levelsandcopynumberburdeninapopulationbasedcase‐controlstudy,ChildhoodAutismRisks
fromGeneticsandEnvironment(CHARGE).Asignificantandsizableinteractionwasfoundbetween
duplicationburdenandozoneexposure(OR2.78,P<0.005),greaterthanthemaineffectforeither
copynumberduplication(OR2.41,95%CI:1.36~4.82)orozonealone(OR1.19,95%CI:
0.75~1.89).Theoverallimplicationofourfindingsisthatsignificantgene‐environmentinteraction
associatedwithautismexistsandcouldaccountforaconsiderablelevelofheritabilitynotdetected
byevaluatingDNAvariationalone.
Categories: CopyNumberVariation,Gene‐EnvironmentInteraction
P132
GeneRegulatoryNetworkinferenceviaConditionalInferenceTreesand
Forests
KyryloBessonov1,FrancescoGadaleta1,KristelVanSteen1
1UniversityofLiege
Treesareclassicaldatastructuresallowingeffectivelyclassifyingandpredictingresponses.Dueto
versatilityandhighperformanceinclassificationandprediction,thereexistplentyoftree‐based
methodsincludingpopularConditionalInferenceTree(CIT)andForests(CIF),RandomForests
(RF),RandomizedTrees(RT),randomizedC4.5,etc.Inthisworkweassessedtheperformanceof
CITandCIFmethodsincorrectgeneregulatorynetwork(GRN)predictionfromexpressiondataby
usingreferencegoldenstandardbuiltfromrealtranscriptionalregulatorynetworkofE.coli.The
syntheticmicroarrayexpressiondatawasobtainedfromDREAM4challenge.Theperformanceof
eachnetworkinferencemethodwasassessedviaAreaUnderReceiverOperatingCharacteristic
(AUROC)andAreaUnderPrecisionRecall(AUPR)metrics.OurpreliminaryresultsshowthatCIT
andCIFsuccessfullypredictdirectedGRNsatacceptableperformanceratesalthoughnotoptimal
(thebestAUROCat0.68andAUPRat0.13forCIFandthebestAUROCat0.58andAUPRat0.18for
CIT).Surprisinglybyusingthecurrentaggregationschemeoffeatureimportancethatprefers
featureswiththehighestnumberofobservations,asingleCITwasabetterperformercomparedto
CIFsinall5networks.Nevertheless,theCIFsshowedanoverall10%improvementinAUROC.A
singleCIThas24%andCIFshave27%loweroverallperformancecomparedtothebestperformer
ofDREAM4ChallengebasedoncumulativeareasofPRandROCcurves.Weplantotestother
featureimportanceaggregationtechniquesinasingletreeandintreeensemblesinorderto
outperformthetopDREAM4algorithms.Inadditiontheeffectsofexpressiondatastandardization
tounitvariancewillbepresented.Infuture,thedevelopedCIFframeworkwillbeusedtoperform
dataintegrationanalysisofmulti‐omicsdatasets.
Categories: DataIntegration,Gene‐GeneInteraction,GeneExpressionArrays,GeneExpression
Patterns
P133
PREDICTINGTHEGENETICRISKFORCOMPLEXDISEASES:CHOOSING
THEBESTPOLYGENICRISKSCOREFORTYPEIIDIABETES
KristiLäll1,2,KristaFischer1,ReedikMägi1,TõnuEsko1
1EstonianGenomeCenter,UniversityofTartu
2InstituteofMathematicalStatistics,UniversityofTartu
Weassessthepracticalvalueoftheresultsfromlarge‐scalegenome‐wideassociationstudies
(GWAS)inpersonalisedriskpredictionforType2Diabetes(T2D).Alargenumberofassociated
variants(SNPs)acrossthegenomehasbeenidentified,eachhavingarelativelyweakeffectonthe
T2Drisk.Thismotivatestheuseofpolygenicriskscores,definedasweightedsumsofriskallele
frequencies.Wediscussdifferentoptionsofconstructingsuchscoresinpracticeandstudytheir
advantagesanddisadvantages.Themainselectioncriterionforamarkertobeincludedinthescore,
isitssignificance(p‐value)intheGWASmeta‐analysis,whereastheestimatedlogisticregression
coefficientsareusedasweights.Mostoften,onlythegenome‐widesignificantmarkers(p<5*10‐8)
areusedinsuchscoresatthemoment.Somestudies,however,proposeincludingalargernumber
ofindependentSNPs,settingthep‐valuethresholdintherange0.1..0.5orincludingallavailable
markers.DifferentversionsofpolygenicriskscoresforTypeIIDiabetes(T2D)willbeconstructed
forthecohortoftheEstonianBiobank.Wewillshowthatincreasingthenumberofmarkersinthe
polygenicriskscoreforT2Dimprovesthepredictiveabilityuntilacertainp‐valuethreshold.In
addition,asignificantinteractioneffectbetweentheoptimalpolygenicriskscoreandBodyMass
Index(BMI)ontheprevalenceofT2Disdetected.BasedonROC‐andreclassificationanalysiswe
concludethatmostadequateriskpredictionshouldaccountforage,BMIandpolygenicriskscore,
whereasthepredictiveabilityofthepolygenicriskscorediffersacrossdifferentBMIcategories.
Categories: Diabetes
P134
Epigenome‐wideassociationwithsolublecelladhesionmoleculesamong
monozygotictwins
YanVSun1,JackGoldberg2,DeanJones3,ViolaLVaccarino1
1EmoryUniversityRollinsSchoolofPublicHealth,Atlanta,GA,USA
2UniversityofWashingtonSchoolofPublicHealth,Seattle,WA,USA
3EmoryUniversitySchoolofMedicine,Atlanta,GA,USA
Inflammationplaysacriticalroleinthepathogenesisofcardiovasculardisease.Epigenetic
mechanisms,includingDNAmethylation(DNAm),havebeenshowntobecriticalintheregulation
ofinflammatorygenes,andcanbeinfluencedbyinflammation.Thesolubleformofcelladhesion
molecules,includingvascularadhesionmolecule1(sVCAM1),intercellularadhesionmolecule1
(sICAM1),andP‐selectin(sP‐selectin),areestablishedbiomarkersforinflammationandendothelial
function,andhavebeenlinkedtocardiovascularevents.
Toidentifyepigeneticmarkersassociatedwithinflammationandendothelialfunction,we
conductedamethylome‐wideassociationstudyofperipheralbloodcellsfrom140monozygotic
(MZ)middle‐agedmaletwinsfromtheEmoryTwinStudy.Usingtworandomlyselectedsubsets
consistingofunrelatedsubjects,weidentifiedandreplicated69and23DNAmsitessignificantly
associatedwithsVCAM1,andsICAM1respectively,adjustedformultipletesting,butnoneforsP‐
selectin.All23sICAM1‐associatedDNAmsiteswerealsoassociatedwithsVCAM1,includingsiteson
genesANKRD11,KDM2B,CAPS,CUX1,andHLA‐DPA1.TwooftheseDNAmsites,locatedonUNC5D
andTMEM125,werealsosignificantcomparingMZtwinswhowerephenotypicallydiscordantfor
bothsICAM1(P=1.79×10‐7,2.78×10‐6)andsVCAM1(P=1.70×10‐9,1.71×10‐7).Theseresults
suggestthatsVCAM1andsICAM1,butnotsP‐selectin,maysharecommonpathophysiologyin
inflammationandendothelialfunctionviaanepigeneticmechanism.Inaddition,theepigenetic
associationwithinflammationcanbedrivenbyunsharedenvironmentalexposures.
Categories: EpigeneticData,Epigenetics
P135
Genapha/dbASM:webbasedtoolstoinvestigateallele‐specific
methylation
GeorgeEllis1,BiLingChen1,KevinUshey1,DeniseDaley1
1UniversityofBritishColumbia
Asinterestinstudyingallele‐specificmethylation(ASM)andit'sassociationwithcommoncomplex
diseasesgrows,thereisaneedforaresourcethatstoresandcatalogsSNPsandregionsthat
demonstrateallelespecificmethylation,analogoustoNCBI'sdbSNP.Additionally,asASMisa
regulatorymechanismthatmaybeassociatedwithhitsfromgenome‐wideassociationstudies
(GWAS),researchersneedasuiteoftoolstohelpthemevaluatetherelationshipbetweenGWAShits
andASM.Tofacilitatetheseinvestigations,wehavecreatedanewwebresourcecalleddbASM,
hostedontheGenaphawebserver(www.genapha.ca).TheaimofdbASMistwofold:1.Curatefrom
theliteratureapublicly‐accessibledatabaseofknownsitesofASM.2.Provideresearcherswitha
web‐basedplatformoftoolsforexploringASManddeterminingregionsofinterest.Wewillpresent
thedbASMresourceincludingdetailsontheunderlyingdatabaseconstructionanddatasets,in
additiontothewebtoolsandexampleworkflows.Thewebtoolsthatarecurrentlyavailableare:
GWASCatalogSNPSearch,ASMSNPSearch,SNPCounter,MethylationPlotsGeneration,and
SequenceViewer.GWASCatalogSNPSearchallowsbrowsingthroughNHGRI'sCatalogofPublished
Genome‐WideAssociationStudiesbyphenotypeandfilteringGWASSNP'sbasedontheirrelationto
suspectedsitesofASM.Forexample,rs11742570isassociatedwithinflammatoryboweldisease
(p=2.0E‐82)anddemonstratesASM.ASMSNPSearchsupportsfindingSNP'sbasedon:ASMstatus
orinterrogability;locationcomparedtogenes,achromosomalregion,orotherSNP's;andfiltering
bypopulationminorallelefrequenciesandsamplesize.SNPCounterusesasynchronousJavaScript
callstothedatabasetoprovidereal‐timecountsoftypesofSNP'sinuser‐selectedregionsof
chromosome.MethylationPlotsGenerationisacalculatesSNPcorrelationstratifyingbygenotype
withCpGsitemethlyationpatterns,similartoepigenomewideassociationstudies(butwithout
diseasestatus),usingCEPHHapMapsamplesandgenotypesandmethylationassaysonthesesame
samplescompletedontheIllumina27Karray.SequenceViewerdisplaysSNP'sinthehuman
referencegenome(basedcurrentlyonGRCh37.p10anddbSNPbuild137)withannotationsshowing
ASMSNP'sandregionsofinterrogabilityviaMSREcutsitesforenzymes:HpyCH4IV,AciI,HhaI,and
HpaI.Thesetoolsareallfreelyavailableforuseat:http://genapha.icapture.ubc.ca/asm/.
Categories: Epigenetics
P136
Agene‐basedmethodforanalysisofIllumina450Kmethylationdata
CeliaMTGreenwood1,2,KathleenKleinOros1AureliaLabbe3,StephanBusche4,JohnLambourne4,
ChristianAPineau5,6,SashaBernatsky5,6,InesColmegna5,6,AntonioCiampi3,TomiPastinen7,Marie
Hudson1,5
1LadyDavisInstituteforMedicalResearch,JewishGeneralHospital
2McGillUniversity,Montreal,QC,Canada
3DepartmentofEpidemiology,BiostatisticsandOccupationalHealth,McGillUniversity
4McGillUniversityandGenomeQuebecInnovationCentre,McGillUniversity
5DepartmentofMedicine,McGillUniversity,Montreal,QC,Canada
6ResearchInstituteoftheMcGillUniversityHealthCentre
7DepartmentofHumanGenetics,McGillUniversity
TheepigeneticeffectsofDNAmethylationplayacriticalroleinregulatinggeneexpressionin
humanhealthanddisease.TheIllumina450Kmethylationarrayallowsthequantificationof
methylationlevelsatover480,000CpGsitesthroughoutthegenome.Thelargenumberofprobes
andtheinherentcorrelationstructureamongnearbyprobesmakeitworthconsideringmultiple‐
probeanalyses.Hereweproposearegion‐basedmethodtoincreasepowerindetectingpatternsof
differentialmethylation,andwecomparetounivariateanalysesinasampleofpatientswith
systemicautoimmunerheumaticdiseasesSARDS).Aspartofanongoingprogramofresearchon
epigeneticsignaturesofSARDS,werecruitedthefollowingsubjects:seropositiverheumatoid
arthritis(n=12);systemicsclerosis(n=17);andsystemiclupuserythematosus(n=12).Illumina
450Kmethylationdatawasobtainedoncell‐sortedCD4+TlymphocytesandCD14+monocytes
fromallpatientsatbaseline.Similarcellsubsetswereretestedinagroupofpatientsthatreceived
Methotrexatetreatment.Thedatawastransformedusingtwoalternativemethods,alogit
transformationandabetaquantiletransformationtostabilizevariances.Weperformedprobeby
probeunivariatetestsusingbeta‐distributionregressions.Forthegenebasedtests,wefitsparse
principalcomponentmodelsusingallprobeswithin5kbofgeneboundaries.Wethentestedfor
associationbetweenthefirstfewPCsandcelltype/diseasestatus.Gene‐basedanalysismayhave
increasedpowertodetectsubtlechangesinmethylationpatternsacrossgenomicregions.Thisset
ofdataprovidesauniqueopportunitytostudydiseasealterationsinmethylationdataun‐
confoundedbycelltypedifferences.
Categories: Epigenetics,MultivariatePhenotypes
P137
Takeresearchtothenextlevelwithsecondarydataanalyses:Fine‐
mappingthespecificlanguageimpairmentgene
WilliamCLStewart1,ChristopherWBartlett1
1TheResearchInstituteatNationwideChildren'sHospital
MappingthegenemutationsresponsibleformostsimpleMendeliandisorderswasamajorstep
forwardinthefieldsofHuman&MedicalGenetics.However,asincredibleasthemathematicaland
statisticaltoolsthatfacilitatedthisachievementwere,amorepowerfulcollectionofmethodsis
neededtomapthemajorgenesthatinfluencecommon,complexdisease.Tothisend,wedeveloped
whatmaybethemostpowerful,integratedsuiteofstatisticalgeneticssoftwaretodate.Oursuiteis
designedspecificallyforthesecondaryanalysesofexistinggeneticdata,althoughtheanalysisof
newlyacquireddataiseasilyperformed.Themethodswithinoursuiteare(1)optimizedforparallel
computing;(2)rootedinstatisticaltheorywithsubstantialgainsforlargesamples;(3)canintegrate
linkage,case‐control,andfamily‐basedassociationwithgeneexpressiondata;and,(4)interrogate
bothcopynumberandsinglenucleotidevariants.Theresultinghigh‐speed,mathematically
rigorous,andsynergisticcapabilitiesofoursuitearelikelytodefinethenext‐generationofmethods
development.Asaproofofprinciple,weappliedtwoprogramsinoursuite:EAGLETandPOPFAM
tothesecondaryanalysisoffourlargefamiliessegregatingaspecificlanguageimpairmentgeneon
chromosome13.WefoundthatEAGLETreducedthesizeofthecandidateregionby5megabases,
andthatPOPFAM—whichincorporatesinformationfrommatchedcontrolsandreferencessamples,
increasedourabilitytodetectassociatedvariantsbeneaththelinkagepeak.Overall,thisshould
significantlyaidre‐sequencingeffortsaswecloseinonthecausalalleles.
Categories: FineMapping,LinkageAnalysis,LinkageandAssociation,MarkovChainMonteCarlo
Methods,MaximumLikelihoodMethods,MultilocusAnalysis
P138
DetectionofGene‐GeneInteractioninAffectedSibPairsAllowingfor
Parent‐of‐OriginEffects
Chih‐ChiehWu1,SanjayShete2
1NationalChengKungUniversity
2MDAndersonCancerCenter
Genome‐wideassociationstudieshavediscoveredseveralhundredgeneticvariantsassociatedwith
commondiseases,whichinmostsituationsexplainasmallfractionoftheheritability.Gene‐gene
interactionscanplayanimportantroleindiseasesusceptibilityandmayaccountforsomeofthe
missingsusceptibility.Parent‐of‐origineffectsrefertothedifferentialexpressionsofagene
betweentwoparentalchromosomesandhavebeenincreasinglyobservedinmammals.The
developmentofstatisticalmethodsisimportantandneededthatarecapableofcapturingjoint
actionsofindividualgeneticcomponentsunderlyingthediseasesusceptibilityandallowforparent‐
of‐origineffects.Here,weextendedourpreviousallele‐sharingmethodandpresented3
mathematicaltwo‐locusmodelsincorporatingparent‐of‐origineffects:additive,multiplicative,and
generalmodels.Ourmethodsaremodel‐freebasedonallelicidentity‐by‐descentsharingbyaffected
sibpairs.Weproposetheuseoftwo‐locusscoremethodtoassessthegene‐geneinteractioneffects
usingaffectedsibpairsinthepresenceofparent‐of‐origineffects.
Categories: Gene‐GeneInteraction
P139
StudyDesignsforPredictiveBiomarkers
AndreasZiegler1
1UniversityofLübeck,InstituteofMedicalBiometryandStatistics
Biomarkersareofincreasingimportanceforpersonalizedmedicine,includingdiagnosis,prognosis
andtargetedtherapyofapatient.Examplesareprovidedforcurrentuseofbiomarkersin
applications.Itisshownthattheiruseisextremelydiverse,anditvariesfrompharmacodynamicsto
treatmentmonitoring.Theparticularfeaturesofbiomarkersarediscussed.Beforebiomarkersare
usedinclinicalroutine,severalphasesofresearchneedtobesuccessfullypassed,andimportant
aspectsofthesephasesareconsidered.Somebiomarkersareintendedtopredictthelikelyresponse
ofapatienttoatreatmentintermsofefficacyand/orsafety,andthesebiomarkersaretermed
predictivebiomarkersor,moregenerally,companiondiagnostictests.Usingexamplesfromthe
literature,differentclinicaltrialdesignsareintroducedforthesebiomarkers,andtheirprosand
consarediscussedindetail.
Categories: Gene‐EnvironmentInteraction,GeneticDataforClinicalTrialDesign
P140
DoestheFTOgeneinteractwiththesocio‐economicstatusontheobesity
developmentamongyoungEuropeanchildren?ResultsfromtheIDEFICS
study
RonjaForaita1,FraukeGünther1,WenckeGwozdz2,LuciaAReisch2,PaolaRusso3,FabioLauria3,
AlfonsoSiani3,ToomasVeidebaum4,MichaelTornaritis5,IrisPigeot1,onbehalfoftheIDEFICS
consortium
1LeibnizInstituteforPreventionResearchandEpidemiology‐BIPS,Bremen,Germany
2CopenhagenBusinessSchool,DepartmentofInterculturalCommunicationandManagement,Frederiksberg,
Denmark
3NationalResearchCouncil,InstituteofFoodScience,EpidemiologyandPopulationGenetics,Avellino,Italy
4NationalInstituteforHealthDevelopment,DepartmentofChronicDiseases,Tallinn,Estonia
5ResearchandEducationInstituteofChildHealth,Strovolos,Cyprus
Varioustwinstudiesrevealedthattheinfluenceofgeneticfactorsonpsychologicaldiseasesor
behaviorismoreexpressedinsocio‐economicallyadvantagedenvironments.Otherstudies
predominantlyshowaninverserelationbetweensocio‐economicstatus(SES)andchildhood
obesityinwesterndevelopedcountries.TheaimofthisstudyistoinvestigatewhethertheFTOgene
interactswiththesocio‐economicstatus(SES)onchildhoodobesityinasubsampleoftheIDEFICS
cohort(N=4406).Astructuralequationmodel(SEM)isappliedwiththelatentconstructsobesity,
dietaryhabits,physicalactivityandfitnesshabits,andparentalSEStoestimatethemaineffectsof
thelatterthreevariablesandaFTOpolymorphismonobesity.Further,amultiplegroupSEMisused
toexplorewhetheraninteractioneffectbetweenthesinglenucleotidepolymorphismrs9939609
withintheFTOgeneandSESexists.Overallmodelfitwasinconsistent(RMSEA=0.05;CFI=0.79).
SignificantmaineffectsareshownforSES(standardizedβs=‐0.057),theFTOhomozygousrisk
genotypeAA(βs=0.177)andphysicalactivityandfitnesshabits(βs=‐0.113).Theexplainedvariance
ofobesityisabout9%.ThemultiplegroupSEMshowsthatSESandFTOinteractintheireffecton
childhoodobesity(Δχ2=7.3,df=2,p=0.03)insofaraschildrencarryingtheprotectiveTTgenotype
aremoresusceptibletoafavorablesocialenvironment.
Categories: Gene‐EnvironmentInteraction
P141
IdentificationofClustersinNetworkGraphsbyaCorrelation‐based
MarkovClusterAlgorithm
MartinLJäger1,RonjaForaita1
1LeibnizInstituteforPreventionResearchandEpidemiology‐BIPS,Bremen,Germany
Acommongoalingeneexpressionanalysisistoidentifygroupsofgeneswithcorrelating
expressionlevels.TheMarkovClusterAlgorithm(MCL)1isamethodtoidentifysuchclustersin
undirectednetworkgraphs.Itconvertsthegraph’sadjacencymatrixintoaprobabilitymatrixwhich
isthenexpandedandinflateduntilitconverges.Clusterscanbededucedfromtheresulting
equilibriumstatematrix.However,theMCLconsidersassociationsbetweengenesonlyina
dichotomousmanner.Hence,ourobjectiveistoexaminewhethertheMCLbasedonthepartial
correlationidentifiesmorereasonableclusters.Asimulationstudyconsistingofthreedifferently
sizedgeneexpressionnetworksandsixtypesofclustersiscarriedout.Thesetypesofclustersdiffer
insize,numberofclustersexisting,underlyingdistributionandstructure.Eachclustertypeis
modelledinageneexpressionnetworkconsistingof100observationsand100,500and1000
genes,respectively.Weconduct1000replicationsforeachcombinationofclustertypeandnetwork
size.Theperformanceofthepartialcorrelation‐basedMCLiscomparedtotheadjacency‐basedMCL
aswellastok‐meansclusteringandPART(PartitioningAlgorithmbasedonRecursive
Thresholding)2whichareappliedusingthegapstatistic3.TheadjustedRandindex4isusedto
assesstheextenttowhichclustersmatchthetrueclustersandtocomparethealgorithmsamong
eachother.References:[1]VanDongen.PhDthesis,2000,UniversityofUtrecht.[2]Nilsenetal.Stat
ApplGenetMolecBiol,2013,12(5):637‐652[3]Tibshiranietal.JRStatSocB,2001,63(2):411‐
423[4]Hubert&Arabie.JClassif,1985,2(1):193‐218
Categories: GeneExpressionPatterns
P142
Developnovelmixturemodeltoestimatethetimetoantidepressant
OnsetofSSRIsandthetimingeffectsofkeycovariates
YinYao1,MengYuanXu1,WeiGuo1
1NationalInstitutesofMentalHealth
Longitudinaldatasetsondrugonset—whichhaveonlyrecentlybecomeavailableforresearch—
requiremultiple‐pointmeasurements.Wesoughttodevelopastatisticalmodelcapableofanalyzing
longitudinaldatawithtippingpoints—specifically,thepointintimewhenatherapeuticdrugbegins
totakeeffect.Wehavetermedthisnovelmethodthe‘mixturemodel’.Totakeunderlyingdriving
factorsintoaccount,wealsotestedtheassociation(s)betweentimeofonsetandpotential
underlyingfactors.Thenewmixturemodelproposednotonlymodelsadrugonsetbutalsotestsits
associationswithinfluentialvariablessuchasgender,age,anddiseasesubtype.Inordertoestimate
timeofonset,dataweredividedintothreestages:1)drugnaïvestate;2)drugonset;and3)
identifiabledrugeffects.Inadditiontoestimatingwhenonsetoccurs,ourproposedstatisticalmodel
takesintoaccountanyassociationswithpotentiallyinfluentialfactors.Weconductedfour
simulationstudiestotestthefeasibilityofournewmethod,andalsoappliedittoreal‐worlddata
fromtheSTAR*Dstudy.Themixturemodelidentifiedtheeffectofthesedifferentvariablesontime
toonsetofdrugeffects.Whilethelimitedsamplesizemakesitdifficulttogeneralizeany
conclusionsfromthisstudy,severalclinicallyrelevantobservationsemerged.Ourresultsindicated
thatfornon‐anxiousandyoungerpatients,theeffectsofcitalopramwereapparentearlier—bythe
sixthweek;incontrast,forthoseindividualsclassifiedashavinganxietyatbaseline,drugeffectsdid
notappearuntiltheeighthweekoftreatment.
Categories: GeneticDataforClinicalTrialDesign,PredictionModelling
P143
Definingrecombinationhotspotblocks:Justhowhotishot?
Tae‐HwiSchwantes‐An1,HeejongSung1,AlexaJMSorant1,JeremyASabourin1,CristinaMJustice1,
AlexanderFWilson1
1NationalHumanGenomeResearchInstitute/NationalInstitutesofHealth
Inthepastdecade,thenumberofavailablegeneticmarkersusedingeneticstudiesofhuman
diseasehasgrownexponentially.Fromdozensofmicrosatellitesforlinkagestudiestomillionsof
markersinGWASchipsandinwholegenome/exomenext‐generationsequencingincurrent
family/associationstudies,theincreasingdensityofmarkershasbeeninstrumentalforthefine‐
mappingofthehumangenome.However,theincreaseinmarkerdensityhasmadeitincreasingly
difficulttoadjustformultipletestsbecauseofcorrelationsbetweenmarkerscausedbylinkageand
gameticdisequilibrium(LD,GD).Definingandidentifyingtheindependentregionsofthegenome
canprovideanalternativeassessmentofthenumberof“independent”testsfornextgeneration
sequencing.Onemethodthatcanbeusedtoidentifyregionsof“independent”regionsinthe
genomeisbyidentifyingblocksofthegenomethatareflankedbyrecombinationhotspots.
Recombinationhotspotsaredefinedasregionsofthegenomethatshowanincreasedrateof
recombinationthanexpectedatrandom1cM/Mb(1centimorganpermegabase).Theseblockscan
beusedtoidentifyblocksofthegenomethataremostlyindependentfromoneanother.Toidentify
theseindependentblocks(regionsdividedbyrecombinationhotspots),thegenomeisclassified
intohotspots(regionsaboveapredefinedrecombinationthreshold)andcoldspots(regionsbelow
athreshold)usingrecombinationrates(cM/Mb);countsandaveragesizeofthehot/coldspot
blockscanbedetermined.Increasingthethreshold(5%,10%,15%,and20%)increasesthe
averagesizeofthecoldspotsanddecreasesthenumberofhotspots,howevertheaveragesizeof
hotspotsdoesnotappeartochange.
Categories: GenomicVariation
P144
Complexgenealogies,simplegeometricstructures
MarcJeanpierre1
1Université.Descartes
Ancientvariationsalwayshavealongandcomplicatedhistory.Ashaplotypedecayisessentially
stochastic,simplegeometricstructuresthatcanbedescribedunambiguouslyinmathematicalterms
canprovidethealgebraicframeworkforanalysingtheforcesshapingthegenealogyofasingle
allele,oraclusterofvariants.Consideringthesimplestexample,athree‐branchbifurcatingtree,
therearetwopossiblewaysofaddingabranchtoanexistingpairofbranches.Thesetwo
independentandcomplementarypathsofconstructionarerepresentedbytwoalternative
equations.Thepossibleconstructionpathsthereforereflectthehierarchicalorganizationofthe
tree.Star‐likegenealogiesarebyfartheeasiesttoanalyze,asthismodelbypassesallthedifficulties
oftranslatingasetofmosaichaplotypesintoaspecificgenealogy.Innon‐stargenealogies,thereare
alwaysseveralpossiblewaystobreakdownacomplexgenealogyinsubtrees.Thedifferent
constructionpathsrepresentingalternativesequenceofeventsthatmaybeobservednaturally
makesuseofparametersasbranchlengthsthatcanberepresentedgraphically.Subtreesare
conditionallyindependentfromupstreamnodesandequationsrepresentingspecificsequencesof
eventsmaybeconstructedfrombottomtotop.Theshapeofthetreeneededtodecipherthehistory
ofmutationancestryisamathematicalabstraction.Thedefinitionofhaplotypeblocksasphysical
entities,withclearborders,asforobjectsinthephysicalworld,resultsinanapparent
simplification,butisnotreallyhelpfulbecauseunnecessaryreductionofcomplexitypreventsthe
derivationofmeaningfulpatterns.
Categories: HaplotypeAnalysis,MultipleMarkerDisequilibriumAnalysis
P145
Missingheritabilitypartiallyexplainedbysequentialenrollmentofstudy
participants
DamiaNoce1,MartinGögele1,ChristineSchwienbacher1,AlessandroDeGrandi1,YuriD'Elia1,PeterP
Pramstaller1,CristianPattaro1
1CenterforBiomedicine,EuropeanAcademyofBolzano/Bozen(EURAC)(affiliatedInstituteoftheUniversity
ofLübeck),Bolzano,Italy
Inpedigree‐basedstudiestherecruitmentstrategycouldplayanimportantroletoexplainpartof
themissingheritability.Arecruitmentcarriedonoveralongtimeperiodmightpairupwith
seasonalorday‐specificconditions,suchasambienttemperature,sampletransportconditionsand
laboratorysamplehandling,introducingsamplestratificationsimilartothesibship(SS)effect.To
quantifytheimpactofsuchissues,weanalyzed54bloodparametersfromthefirst2948
participantsoftheCooperativeHealthResearchinSouthTyrol(CHRIS)study,enrolledfromAug
2011untilJul2013andconnectedthroughanextendedpedigree.Tomaximizeparticipationof
completefamiliesweenrolledpreferentiallycloserelativeswithinthesameday(up10perday).
Geneticheritability(h2)wasestimatedbyfittingsex‐andage‐adjustedvariancecomponents
models.Weadditionallyincludedsharedenvironmentaleffectsdefinedasdayofparticipation
(DoP),dailytemperature(DT)andSS.Weobservedah2reductionfor49,28and39traitswhen
accountingforDoP,DTandSS,respectively.WhenincludingtheDoP,theh2reductionwas>10%
for11traitsand>40%forsodium,chlorine,calciumandmeancorpuscularhemoglobin
concentration.TheSSeffectinduced>10%h2reductionfor10traitsand>40%onlyforcortisol.
Despitebeingassociatedwithsometraits,DTdidnotalterh2estimatessubstantially.Thedayof
participation,asaproxyforissuesthatmayhappenduringtheenrollmentormeasurementphase,
canbeanimportantstratificationfactor,whichmayinducestrongerheritabilityoverestimation
thanthesibshipeffect.Whenappropriate,itshouldbeusedtocomplementthesibshipeffectto
preventpopulationstratification.
Categories: Heritability
P146
RobustPrincipalComponentAnalysisAppliedtoPopulationGenetics
Processes
CarineLegrand1,JustoLorenzoBermejo1
Ithasbeenshownthatprincipalgeneticcomponentsreflectevolutionaryprocessesandthegenetic
parametersofapopulation.Forexample,McVeanprovidedagenealogicalinterpretationof
principalcomponentanalysis(PCA)[1].WehaveexaminedtheabilityofseveralrobustPCA
methodstomirrorpopulationchangesinheterozygosityandadmixture.Evolutionaryprocesses
weresimulatedusingsimuPOPandownscripts[2].Wefirstexaminedgeneticdriftinasingle
population(CEUhaplotypesfromHapMap)consideringgrowth,recombinationandselection.We
alsosimulatedgeneflowinSouthAmericaafterthearrivalofindividualswithEuropeanandAfrican
ancestries,allowingforafastpopulationgrowthinthelastcentury.CEUandYRIsamplesfromthe
1000GenomesProjectrepresentedEuropeanandAfricancomponents.PCAresultsmotivatedthe
useofMXL(Mexican)insteadofCLM(Colombian)genotypesassurrogatesofnativeSouth
Americanancestry.Heterozygosityandadmixturewerequantifiedintheevolvingpopulations,and
theirrelationshipwiththeprincipalgeneticcomponentsestimatedbystandardPCA,sphericalPCA,
andminimumcovariancedeterminantmethodswasexamined.Resultsfromthegeneticdrift
scenariorevealedastrongercorrelationbetweenheterozygosityandtherobustprincipalgenetic
components.ThesimulationofSouthAmericanadmixturealsorevealedapotentialadvantageof
robustPCA.Resultsfromongoingsensitivityanalyseswillbepresentedattheconference.
[1]McVeanG(2009)AGenealogicalInterpretationofPrincipalComponentsAnalysis.PLoSGenet
5(10):e1000686.
[2]PengB,KimmelM(2005)simuPOP:aforward‐timepopulationgeneticssimulation
environment.Bioinformatics,21(18):3686‐7.
Categories: Heterogeneity,Homogeneity,PopulationGenetics,PopulationStratification
P147
Identifyingfoundersmostlikelytohaveintroduceddisease‐causing
mutationswiththeRpackageGenLib
ClaudiaMoreau1*,Jean‐FrançoisLefebvre1*,HéloïseGauvin1,2,MichèleJomphe3,ChristophPreuss1,
GregorAndelfinger1,4,DamianLabuda1,4,HélèneVézina3,Marie‐HélèneRoy‐Gagnon1,5
*Theseautherscontributedequallytothiswork
1CHUSainte‐JustineResearchCenter,Montreal,Quebec,Canada
2DepartmentofSocialandPreventiveMedicine,UniversitédeMontréal,Montreal,Quebec,Canada
3BALSACProject,UniversitéduQuébecàChicoutimi,Chicoutimi,Quebec,Canada
4DepartmentofPediatrics,FacultyofMedicine,UniversitédeMontréal,Montreal,Quebec,Canada
5DepartmentofEpidemiologyandCommunityMedicine,UniversityofOttawa,Ottawa,Ontario,Canada
Founderpopulations,suchastheFrenchCanadian(FC)populationofQuebec,Canada,playan
importantroleinthestudyofgeneticdiseases.Theiradvantagesoftenincludeaccesstodetailed
genealogicalrecords.Onceevidenceforadisease‐causingmutationhasbeenfound,genealogical
datacanbeusedtoidentifythefoundersmostlikelytohaveintroducedthemutationinthe
population.Largegenealogicaldatarequirespecializedanalyticalmethodsandsoftware.We
presenttheRpackageGenLibforgenealogicalanalysis.GenLibcancomputerelevantsummary
measuresdescribinggenealogiesandrelatedness,includingkinshipandinbreedingcoefficients.It
alsoperformsgene‐droppingsimulations.Inthisstudy,weextendedtheGenLibgene‐dropping
simulationfunctiontotakeintoaccountthelengthofthesegmentpassedIBDthroughgenerations
andafitnessparameterforhomozygotes.Thisextensionallowsamorepreciseestimationthrough
simulationsoftheprobabilitythatthesharedsegmentdescendedfromaspecificfounder.We
illustratetheuseofGenLibwithgenealogicaldatafrom11patientswiththerecentlyidentified
autosomalrecessivesyndromeofChronicAtrialandIntestinalDysrhythmia(CAID).Average
kinshipandinbreedingcoefficientsofthesepatientswere0.002and0.004,respectively.Wefound
thatonefoundingcouplehadaprobabilityover80timeslargerthanthatofanyotherfoundersto
haveintroducedthemutationintheFCpopulation.ThiscoupleimmigratedtoQuebecCityfrom
Francearound1621.Theseresultsprovideinformationonexpectedfrequenciesofthediseasein
thepopulationandonthediffusionpatternofthemutationontheQuebecterritory.
Categories: Inbreeding,IsolatePopulations,PopulationGenetics
P148
RegionalIBDAnalysis(RIA):linkageanalysisinextendedpedigrees
usinggenome‐wideSNPdata
JakrisEu‐ahsunthornwattana1,2,HeatherJCordell1
1InstituteofGeneticMedicine,NewcastleUniversity,InternationalCentreforLife,CentralParkway,Newcastle
uponTyne,NE13BZ,UK
2DivisionofMedicalGenetics,DepartmentofInternalMedicine,FacultyofMedicineRamathibodiHospital,
MahidolUniversity,RamaVIRd,Ratchathevi,Bangkok10400,Thailand
Exactcalculationsfortraditionallinkageanalysisarecomputationallyimpracticalinlarge,extended
pedigrees.Althoughsimulation‐basedmethodscanbeused,theyarenotexactandstillrequire
significantcomputationalwork.Forthesecircumstances,weproposeRegionalIBDAnalysis(RIA),a
non‐parametriclinkagemethodbasedoncomparisonoflocallyandgloballyestimatedidentityby
descent(IBD)sharinginaffectedrelativepairs.Inthismethod,genome‐wideSNPdataareusedto
calculatethe“global”expectedIBDsharingprobabilitiesspecifictoeachaffectedrelativepair,
againstwhicha"local"setofIBDsharingprobabilities,estimatedusingSNPdatawithinawindowof
pre‐specifiedwidth,canbecompared.TheseIBDsharingprobabilitiescanbeestimatedusinga
varietyofprograms/methods:weusedPLINKandKINGinthisstudy.TheglobalandlocalIBD
sharingprobabilitiescanbeusedtoconstructanon‐parametricmaximumlikelihoodstatistic
(MLS)‐liketestoflinkageineachwindow.Weillustratetheuseofourmethodtodetectlinkage
signalsinrealnuclear‐familydataandinsimulateddatabasedonlargeextendedpedigrees.This
methodshouldbeusefulinstudiesinvolvinglargeextendedfamilies,withanadditionaladvantage
ofnothavingtorelyonanypriorknowledgeaboutfamilialrelatedness.
Categories: LinkageAnalysis
P149
Polygenicriskpredictionmodelinginpedigreesimprovespower
JefferyStaples1,ChadDHuff2,JenniferEBelow3
1TheUniversityofWashington
2TheUniversityofTexasMDAndersonCancerCenter
3TheUniversityofTexasHealthScienceCenter
Asanalysesofsequencedatainlargepopulationbasedcohortsstruggletoachievesufficientpower
todetectevenlargesignalsfromveryrarevariation,thefamily‐basedlinkageapproachhascome
backintovogue.Inthecontextofcomplexdiseasetraitshowever,abilitytodetecttruesignalis
impededbymodifyingenvironmentalandgeneticfactorsthatinfluenceratesofpenetranceand
phenocopies.Classically,knownmodifiersofdiseaserisk,e.g.age,havebeenmodeledinliability
classes.TheeraofGWAShastaughtusagreatdealaboutcommonunderlyinggeneticeffectson
complextraits.Thesepolygeniceffectsimpactriskofdiseaseandcanactasmodifyingfactorsto
rarevariationsegregatinginpedigrees.Weshowthatmodelingtheseeffectsimprovestheabilityto
bothdetecttruelinkagesignalsoflargeeffectrarevariantsfromthegenomeandcorrectlyidentify
unlinkedmarkers.Insimulationsofgenotypesfora1000differentpedigrees(meansize25,36%
missingsamples)wemodeledphenotypesbymodifyingtheprobabilityofdiseasegivenalarge
effectdominantriskallele,A,usingasimulatedaggregatepolygenicriskscore(pgrs)calculated
from100differentcommonvariants:P(d|aa)=pgrs,P(d|Aa,AA)=pgrs+0.9.WecomparedLOD
scoresatthecausalvariantandanunlinkedvariantwhenmodelingthepgrsinanindividual‐
specificliabilityclasstoscoresderivedfromasinglesharedliabilityclass.Powertodetectthe
casualvariantincreasedin>60%ofoursimulations,overallaveraging>10%increaseandmean
LODgainof>0.25.IncorporatingpolygenicriskpredictionslightlyloweredLODatunlinked
markers.Accuratemodelingofestablishedpolygenicriskfactorsimprovespowerestimatesin
linkagestudies.
Categories: LinkageAnalysis,LinkageandAssociation
P150
Performanceoflinkageanalysisconductedwithwholeexome
sequencingdata
SimonGosset1,EdgardVerdura2,FrançoiseBergametti2,StephanieGuey2,ElisabethTournier‐
Lasserve2,StevenGazal3
1INSERMU1137,IAME,UniversitéParisDiderot,Paris,France
2INSERMU1161,UniversitéParisDiderot,Paris,France
3AssistancePubliquedesHopitauxdeParis(APHP),Paris,France
IdentificationofcausalvariantsinMendeliandisorderwasusuallydonebycombininglinkage
analysis(LA)onlargefamiliesandpositionalcloning.Theprogressofthehigh‐throughput
sequencingledteamstoperformdirectlywholeexomesequencing(WES)fortheidentificationof
thesevariants,particularlyforsinglesmallfamiliesthatcanbeanalysedbyasimplefilteranalysis.
However,itisessentialtominimizethenumberofcandidatevariantsbeforestartingstudieson
theirfunctionalconsequences.Toreducethenumberofvariantsthataresequencingerrors,not
coveredinoneindividual,orwithoutallelicfrequencyinreferencedatabase,andtofacilitatethe
studyofrecessivediseaseswithallelicheterogeneity,anadditionalLAcanbeperformed.Many
studieshavethuscombinedtheirWESfilteringwithaLAonmicrosatellitesorSNPchips,which
uniformlycoverthegenome.PerformaLAoncommonpolymorphismspresentinWESdata
appearsasanattractivestrategytoreducethecostoftheanalyses.However,ithasbeenrarely
done,duetothenon‐uniformexoncoverageofthegenome,andtothelackofknowledgeofLA
poweronthiskindofdata.OurgoalwastostudytheperformanceofLAconductedwithexome
genotypes.Toachievethis,weperformedasimulationstudyof2families(onewithadominant
disease,onewitharecessivedisease)andcomparedLAresultsonWESgenotypesanddatafrom
SNPchips.OurresultsshowthataLAconductedonWESgenotypesexcludesaccuratelyahigh
proportionofthegenome.Inaddition,itsfalsepositiveandfalsenegativeevidenceoflinkagearein
thesamerangethattheonesofLAconductedonSNPchips.Finally,anapplicationonrealdatawill
illustratethebenefitsofthisstrategy.
Categories: LinkageAnalysis,SequencingData
P151
Useofexomesequencingdatafortheanalysisofpopulationstructures,
inbreeding,andfamiliallinkage
VincentPedergnana1,2,AzizBelkadi1,AvinashAvinash3,QuentinVincent1,YuvalItan4,Bertrand
Boisson4,Jean‐LaurentCasanova1,5,LaurentAbel1,5
1LaboratoryofHumanGeneticsofInfectiousDiseases,NeckerBranch,INSERMU1163,UniversityParis
Descartes,ImagineInstitute,Paris,France
2WellcomeTrustCentreforHumanGenetics,Oxford,UnitedKingdom
3NewYorkGenomeCenter,NewYork,NY,USA
4St.GilesLaboratoryofHumanGeneticsofInfectiousDiseases,RockefellerBranch,theRockefellerUniversity,
NewYork,NY,USA
5St.GilesLaboratoryofHumanGeneticsofInfectiousDiseases,RockefellerBranch,theRockefellerUniversity
Numerousmethodshavebeenproposedtoanalyzewholeexomesequencing(WES)datainorderto
discoverpotentialcausalvariantsinMendeliandisordersandinmorecomplextraits.These
methodscouldbenefitfromadditionalinformationsuchaslinkagestudiesinthestudyofMendelian
diseases.PopulationstratificationcouldalsobeanissueintheanalysisofWESdatawhenfocusing
oncomplextraits.Bothlinkageandpopulationstructureanalysesareclassicallyconductedthrough
genome‐wide(GW)SNParrays.Here,wecomparedtheinformationyieldedbyWESdatatothat
providedbySNParraydataintermsofanalysesusuallyperformedbySNParraydatasuchas
principalcomponentanalyses(PCA),linkagestudies,andhomozygosityrateestimation.We
analyzed123subjectsoriginatingfromsixworldregions,includingNorthAfricaandMiddleEast
whichareregionspoorlycoveredbypublicdatabaseandpresentingahighconsanguinityrate.A
numberofqualitycontrol(QC)filtersweretestedandappliedtotheWESdata.Comparedtoresults
obtainedwithSNParraydata,wefoundthatWESdataprovidedaccuratepredictionofpopulation
substructureandledtohighlyreliableestimationofhomozygosityrates(correlation>0.94withthe
estimationsprovidedbySNParray).Linkageanalysesshowedthatthelinkageinformationprovided
byWESdatawasonaverage53%lowerthantheoneprovidedbySNParrayattheGWlevel,but
58%higherinthecodingregions.Inconclusion,WESdatacouldbeusedafterappropriateQCfilters
toperformPCAanalysisandadjustforpopulationsubstructure,toestimatehomozygosityrates,
andtoperformlinkageanalysesatleastincodingregions.
Categories: LinkageAnalysis,PopulationGenetics,SequencingData
P152
FastlinkageanalysiswithMODscoresusingalgebraiccalculation
MarkusBrugger1,2,KonstantinStrauch1,2
Maximilians‐Universität,Munich,Germany
2InstituteofGeneticEpidemiology,HelmholtzZentrumMünchen,GermanResearchCenterfor
ObjectiveThemodeofinheritanceisoftenunknownforcomplexdiseases.Inthecontextof
parametriclinkageanalysis,thisimpliesthataMOD‐scoreanalysis,inwhichtheLODscoreis
maximizedwithrespecttothetrait‐modelparameters,canbemorepowerful.Becausethe
calculationofthedisease‐locuslikelihoodforeverytestedsetoftrait‐modelparametersisthemost
time‐consumingstepinaMOD‐scoreanalysis,weaimedtooptimizethispartofthecalculationto
speed‐uplinkageanalysisusingtheGENEHUNTER‐MODSCOREsoftwarepackage.MethodsOurnew
algorithmisbasedonminimizingtheeffectivenumberofinheritancevectorsbycollapsingthem
intoclasses.Tothisend,thedisease‐locus‐likelihoodcontributionofeachinheritancevectoris
representedandstoredinitsalgebraicformasasymbolicsumofproductsofpenetrancesand
disease‐allelefrequencies.Simulationsofdatasetswereusedtoassessthespeed‐upofournew
algorithm.ResultsFocusingonMOD‐scoreanalysisofsingledatasets,wewereabletoobtainspeed‐
upsrangingfrom1.94foraffected‐sibpairsto11.52foraffected‐sibsextetscomparedtothe
originalGENEHUNTER‐MODSCOREversion.Whenincludingsimulationstocalculateempiricalp
values,thespeed‐uprangedfrom1.69to10.36.Speed‐upwasgenerallyhigherforlargerpedigrees.
ConclusionsComputationtimesforMOD‐scoreanalysisincludingp‐valuecalculationhavebeen
prohibitivelyhighsofar.Withournewalgebraicalgorithm,theevaluationofmanytestedsetsof
trait‐modelparametersduringthemaximizationinaMOD‐scoreanalysisisnowfeasiblewithina
reasonableamountoftime,evenwhenempiricalpvaluesarecalculated.
Categories: LinkageAnalysis
P153
Fetalexposuresandperinatalinfluencesontheprematureinfant
microbiome
DianaAChernikova1,DevinCKoestler2,AnneGHoen3,MollyLHousman4,PatriciaLHibberd5,Jason
HMoore6,HilaryGMorrison7,MitchellLSogin7,MuhammadZUl‐Abideen8,JulietteCMadan9
1DepartmentofGenetics,GeiselSchoolofMedicineatDartmouth
2DepartmentofBiostatistics,UniversityofKansasMedicalCenter
3DepartmentofCommunityandFamilyMedicine,GeiselSchoolofMedicineatDartmouth
4DepartmentofMicrobiologyandImmunology,GeiselSchoolofMedicineatDartmouth
5DepartmentofPediatrics,MassachusettsGeneralHospital
6InstituteforQuantitativeBiomedicalSciences,GeiselSchoolofMedicineatDartmouth
7JosephineBayPaulCenter,MarineBiologicalLaboratory
8GeiselSchoolofMedicineatDartmouth
9DepartmentofPediatrics,Dartmouth‐HitchcockMedicalCenter
Theimpactofmaternalcomplicationsontheprematureinfantmicrobiomeisstilllargely
unexplored.Toinvestigatetheeffectsofthesecomplicationsonthegutmicrobiome,wecollected
serialstoolsamplesobtainedweeklyfromextremelyprematureinfantsenrolledinaprospective
longitudinalstudyfrombirththroughhospitaldischarge,andthensequencedtheV4V6regionof
bacterial16SrRNAgenes.Perinatalmaternalcomplicationsevaluatedincludedprolongedpreterm
prematureruptureofmembranes(PPPROM),chorioamnionitis,deliverymode,andperipartum
antibiotics.Subjectswithprenatalexposuretoanon‐sterileintrauterineenvironment(PPPROMand
chorioamnionitis)werefoundtohaverelativelyhigherabundanceofknownpathogenicbacteria
acrossalltimepointscomparedtosubjectswithoutthoseexposures,irrespectiveofexposureto
postnatalantibiotics.ComparedwiththosedeliveredbyCesareansection,vaginallydelivered
subjectswerefoundtohaveasignificantlylowermicrobialdiversityacrossalltimepoints,with
lowerabundanceofmanybacterialgenera,mostlyinthefamilyEnterobacteriaceae.Hierarchical
clusteringanalysisshowedthatsamplesassociatedwithanon‐sterileuterineenvironment
clusteredtogetherandhadanenrichmentofpathogens;furthermore,thecluster’saverage
microbialdiversityscorethatwassignificantlylowerthanthatofaclusterofsampleswithoutthe
exposure,whichinsteadhadanenrichmentofimportantgutcommensals.Ourresultsdemonstrate
thatexposuretoprenatalpathogensimpactsthedevelopmentoftheprematuregutmicrobiome,
andhighlightsopportunitiestointerveneviabreastmilkfeedings,alteredantibioticregimens,or
probiotics.
Categories: MicrobiomeData
P154
Combininggenotypewithallelicassociationasinputforiterative
pruningprincipalcomponentanalysis(ipPCA)toresolvepopulation
substructures
KridsadakornChaichoompu1,2,RamounaFouladi1,2,PongsakornWangkumhang3,AlisaWilantho3,
WanwisaChareanchim3,SissadesTongsima3,AnavajSakuntabhai4,KristelVanSteen1,2
1SystemsandModelingUnit,MontefioreInstitute,UniversityofLiege,Belgium
2BioinformaticsandModeling,GIGA‐R,UniversityofLiege,Belgium
3BiostatisticsandinformaticsLaboratory,GenomeInstitute,NationalCenterforGeneticEngineeringand
Biotechnology,Thailand
4FunctionalGeneticsofInfectiousDiseasesUnit,InstitutPasteur,France
SingleNucleotidePolymorphisms(SNPs)arecommonlyusedtocapturevariationsbetween
populationsandoftengenome‐wideSNPdataareprunedbasedonlinkagedisequilibrium(LD)
patterns.Notably,haplotypecompositionandthepatternofLDbetweenmarkersmayvarybetween
largerpopulationsbutmayalsoplayarolewithinmoreconfinedgeographicregions.Indeed,
knowledgeabouthaplotypesinunrelatedindividualscanrevealusefulinformationaboutgenetic
ancestry.Here,weuseiterativepruningprincipalcomponentanalysis(ipPCA)[Intarapanich2009]
toidentifyandcharacterizesubpopulationsinanunsupervisedwayusingarichsetofgenetic
markerssinceusingreducedsetsofgeneticmarkersforthesepurposescanbecomechallenging,
especiallywhensimilargeographicregionsareinvolvedorwhenspuriouspatternsarelikelyto
exist.Asinputdata,eitherprunedgenome‐wideSNPdataareusedormultilocushaplotype
informationderivedfromthegenome‐wideSNPpanel.Theseapproachesareappliedtoreal‐life
datafrom4028Vietnameseindividuals[Khor2012].PreliminaryresultsindicatethatipPCAapplied
toprunedSNPdataoripPCAthatexplicitlyusesmultilocusinformation(haplotypes)give
complementaryinformationaboutpopulationsubstructureforgeographicallyconfinedpopulations.
Bothmethodsaddressdifferentaspectsofpopulationstructure.Inconclusion,weproposeto
combineanLD‐basedhaplotypeencodingschemewiththeipPCAmachinerytoretrievefine
populationsubstructures.Despitethecomplexitiesthatareassociatedwithhaplotypeinference,
addedvaluecanbeobtainedwhentheLDstructurebetweenSNPsisexploitedinthesearchfor
relevantpopulationstrata.
Categories: PopulationGenetics,PopulationStratification
P155
Spuriouscrypticrelatednesscanbeinducedbypopulationsubstructure,
populationadmixtureandsequencingbatcheffects
DiZhang1,ShuweiLi1,GaoTWang1,SuzanneMLeal1
1CenterforStatisticalGenetics,BaylorCollegeofMedicine
Itisimportanttoidentifycrypticallyrelatedindividualsinpopulation‐basedassociationstudies,
sinceinclusionofrelatedindividualscanincreasetypeI&IIerrors.Toresolvethisproblemmixed
modelshavebeenproposed,buttheycanbecomputationallyintensiveandtypeI&IIerrorscanbe
inflated.Anotheroptionistoremoverelatedindividualsfromanalysis.Dataqualitycontrolshould
includeidentificationofcrypticallyrelatedindividuals.Cautionshouldbeused,sincepopulation
substructure/admixtureandsequencedatabatcheffectscancausedetectionofspurious
relatedness.Inordertoinvestigatetheproblemweevaluatedtherelatednessof1,092samplesin
1000Genomesand2,300African‐AmericansubjectsfromtheNHLBI‐ExomeSequencingprojectvia
twopublishedmethodsforkinshipinference:(i)thePLINKalgorithmwhichisbasedonidentical‐
by‐descentstatisticundertheassumptionofhomogeneouspopulation,and(ii)theKING‐robust
algorithmwhichusesanestimateofthegenome‐wideaverageheterozygosityacrossindividualsto
computeanestimatorofkinshipcoefficient.Weidentifiedspuriousrelatednessduetopopulation
substructure/admixtureandbatcheffectswithbothmethods,buttheproblemwasmoreseverefor
PLINK.Anexcessof3rddegreerelativeswasobservedduepopulationadmixture/substructureand
batcheffects.Thekinshipcoefficientsalsovarieddependingonhowtheanalysiswasperformedand
individualswerereclassified,e.gfrom1stdegreeto2nddegreerelatives.Inadditiontopresenting
theresultsoftheseanalysesandshowingtheseverityofthebiasesinthekinshipcoefficients,we
alsodemonstratestrategiestoavoidthedetectionofspuriousrelatedness.
Categories: PopulationGenetics,PopulationStratification,SequencingData
P156
Effectofpopulationstratificationonvalidityofacase‐onlystudyto
detectgene‐environmentinteractions
PankajYadav1,SandraFreitag‐Wolf1,WolfgangLieb2,MichaelKrawczak1
1InstituteforMedicalInformaticandStatistic,Christian‐AlbrechtsUniversity,Kiel,Germany
2InstituteofEpidemiology,Christian‐AlbrechtsUniversity,Kiel,Germany
Gene‐environment(G×E)interactionstudiesareassumedtopartiallyfillthegapbetweenthe
estimatedheritabilityofcommonhumandiseasesandthegeneticcomponenthithertoexplainedby
disease‐associatedvariants.Thecase‐only(CO)studyhasbeenproposedasavalidapproachwith
increasedstatisticalefficiencyovercase‐controlandcohortstudiesindetectingG×Einteractions.
However,hiddenstratificationinthestudypopulationcanseverelycompromiseaCOstudy.Noneof
thepriorliteratureexplicitlyaddressedtheeffectofstratificationonaCOstudy.Wetherefore
systematicallyassessedthroughsimulationstheeffectofpopulationstratification(PS)onthe
validityofaCOapproachinG×Einteractionsstudies.Oursimulationsshowthat,whenstudysample
isdividedbybothgeneticandexposurefactors,aCOstudyprovidesaninflatedtypeIerrorrate.
Further,oursimulationsshowthattransmissiondisequilibriumtest(TDT)isrobustagainstgenetic
and/orexposurestratificationindetectingG×Einteractions.
Categories: PopulationStratification
P157
Anovelriskpredictionalgorithmwithapplicationtosmoking
experimentation
RajeshTalluri1,AnnaWilkinson2,MargaretSpitz3,SanjayShete1
1TheUniversityofTexas,M.D.AndersonCancerCenter
University of Texas School of Public Health
2
3BaylorCollegeofMedicine
Riskpredictionmodelsarebeingdevelopedtopredicttheriskofavarietyofcancers,and
cardiovasculardiseases.However,standardapproachesdonotaccountforthevariabilityassociated
withthecohortbeingarandomsamplefromthepopulation.Wedevelopedanovelriskprediction
approachcalledResampling‐basedModelSelectionandAggregationtocomputeabsoluterisk.Our
approachaccountedforvariabilityinthesampledcohortbyresamplingthedataandaggregating
theparameterestimatesfortheresampleddatasets.Wethenusedaresampling‐basedmodel
selectionalgorithmtoselectthepredictorstoincludeinthefinalmultivariableriskmodel.This
approachguardsagainstover‐fittingthemodelandreducesthevarianceofthemodelparameters.
Theperformanceoftheriskpredictionmodelwasevaluatedusingtheareaunderthereceiver
operatingcharacteristiccurve(AUC).Usingtheriskpredictionmodel,wecomputedtheabsolute
riskofsmokingexperimentationinMexicanAmericanyouth.Thedataincludedgeneticandnon‐
geneticfactorsthatwerecollectedatbaseline.TheproposedriskpredictionmodelhadanAUCof
0.719(95%confidenceinterval,0.637to0.801)forpredictingabsoluteriskforsmoking
experimentationwithin1year.
Categories: PredictionModelling
P158
Trio‐BasedWholeGenomeSequenceAnalysisofaCousinPairwith
RefractoryAnorexiaNervosa
PBettyShih1,AshleyVanZeeland2,AndrewBergen3,TristanCarland4,VikasBansal1,Pierre
Magistretti5,WadeBerrettini6,WalterKaye1,NicholasSchork7
1UniversityofCalifornia,SanDiego,LaJolla,CA
2CypherGenomics,LaJolla,CA
3SRI,PaloAlto,CA
4TheScrippsResearchInstitute,LaJolla,CA
5ÉcolePolytechniqueFédéraledeLausanne,Lausanne,Switzerland
6UniversityofPennsylvania,Philadelphia,PA
7J.CraigVenterInstitute,LaJolla,CA
AnorexiaNervosa(AN)hasanonsetduringadolescenceandischaracterizedbyemaciation,fearof
gainingweightdespitebeingunderweight,andhasthehighestmortalityrateofallpsychiatric
illnesses.Despitetheserioushealthandpsychosocialconsequencesofthisillness,veryfew
treatmentsareeffectiveatreversingthecoresymptomsofAN.ANishighlyheritableandshowa
homogeneousclinicalpresentationofpersistentfoodrefusalandhighanxietytraits.However,AN
etiologyisbelievedtobeheterogeneousasnomajorsusceptibilitygenehasbeenconsistently
replicatedinmultiplepopulations.ANsymptomsandpersonalitytraitstendtobepresentin
unaffectedfamilymembersofthepatients,suggestingthatcertainsharedgeneticfactorswithin
eachfamilymaycontributetouniquephenotyperiskoftheaffected.Togaininsightsintotherole
“privatevariants”mayplayinANandtomaximizegeneticinformationfromfamilymembersofAN,
hereweleveragedafamily‐basedstudydesigncombinedwithwholegenomesequencingtosearch
forgeneticvariantsthatmayinfluenceANriskinanaffectedcousinpairtogetherwiththeirparents.
Bycapitalizingonthehomogeneityofthediseasepresentationamongthetwocousins,whoboth
haveadiagnosisofrefractoryAN,wereportmethodsbywhichweinterrogatedshared
chromosomalsegmentstransmittedtothemfromtheircommongrandparentsthatcarriedlikely
AN‐relatedfunctionalvariantsinthisfamily.
Categories: PsychiatricDiseases,SequencingData
P159
Powerandsamplesizeformulasfordetectinggeneticassociationin
longitudinaldatausinggeneralizedestimatingequations
GhislainRocheleau1,LoïcYengo2,PhilippeFroguel2
1.UniversitéLille2,Lille,France
2CNRS8199‐InstituteofBiology,PasteurInstitute,Lille,France
Currently,mostgeneticstudiesonlyexploitcross‐sectionaldatatodetectnovelassociations
betweenaSNPandaquantitativetrait,evenifrepeatedlymeasuredoutcomesareavailablefor
analysis.Insteadoffocusingonsomebaselineorsingletimepointmeasurement,itmightbe
desirabletoidentifySNPsassociatedwiththattraitovertime.Onepossibleapproachtomodel
correlatedmeasuresovertimeisthegeneralizedestimatingequations(GEE),especiallyifinterest
liesindetectingthemeandifferencesofthetraitasafunctionofthegenotypes.Unlikelinearmixed
models,GEEmodelsdonotrequirethejointdistributiontobefullyspecified,onlythemeanandthe
variancemustconformtolinearmodelspecifications,alongwithanappropriatewithin‐cluster
correlationmatrix.However,inpoweranalysis,thiswithin‐clustercorrelationmatrixisoften
unknownandisusuallymodelledasafunctionoftime.Commonchoicesforthismatrixinclude
compoundsymmetry,autoregressive(AR)ormovingaverage(MA)structures.Usingasymptotic
theoryoftheWaldteststatistic,wederiveclosed‐formformulasforpowerandsamplesize
estimationunderanautoregressivemovingaverageARMA(1,1)covariancematrix.Interestingly,
theARMA(1,1)covariancematrixisequivalenttoanAR(1)covariancematrixplusindependent
measurementerror.Weapplyourformulastosimulatedgenotypeandphenotypedata,andtoreal
datacomingfromtheFrenchcohortD.E.S.I.R.(DonnéesÉpidémiologiquessurleSyndrome
d’Insulino‐Résistance).
Categories: QuantitativeTraitAnalysis,SampleSizeandPower
P160
Ontheevaluationofpredictivebiomarkerswithdichotomousendpoints:
acomparisonofthelinearandthelogisticprobabilitymodels
NicoleHeßler1,AndreasZiegler1,2
1InstitutfürMedizinischeBiometrieundStatistik,UniversittzuLübeck,UniversitätsklinikumSchleswig‐
2ZentrumfürKlinischeStudien,UniversitätzuLübeck,Lübeck,Germany
Thestandardstatisticalapproachforanalyzingdichotomousendpointsisthelogisticregression
modelwhichhasmajorstatisticaladvantages.However,someresearcherspreferthelinear
probabilitymodeloverthelogisticmodelinrandomizedtrialsforevaluatingpredictivebiomarkers.
Themainreasonseemstobetheinterpretationofeffectestimatesasabsoluteriskreductionswhich
canbedirectlyrelatedtothenumberneededtotreat.Inthefirstpartofourpresentation,we
provideacomprehensivecomparisonofthetwodifferentmodelsfortheinvestigationoftreatment
andbiomarkereffects.Usingthelogisticregressionmodel,Kraftetal.(2007,HumHered)showed
thatthecombined2degreesoffreedom(2df)gene,gene‐environmentinteractiontestshouldbethe
testofchoicefortestinggeneticeffects.Inthebiomarkertreatmentsettingagenecorrespondsto
thetreatmentandenvironmenttobiomarker.UsingthisanalogyweextendthestudyofKraftetal.
inthesecondpartofourpresentation.Wecompareseveralteststatisticsincludingthe2df
combinationtestusingthelinearprobabilitymodel.Theprosandconsofthecombinedtestare
discussedindetail.Wedemonstratesubstantialpowerlossofthecombinationtestincomparison
witheitherthetestfortreatmentorthetestfortreatment‐biomarkerinteractioninmanyscenarios.
Althoughthecombinationtesthasreasonablepowerinallsituationsconsidered,itspowerloss
comparedtoaspecialized1dftestcanbelarge.Therefore,thecombinedtestcannotbe
recommendedasthestandardapproachinstudiesoftreatment‐biomarkerinteraction.
P161
Atwostagerandomforestprobabilitymachineapproachforepigenome‐
wideassociationstudies
FraukeCDegenhardt1,AndreFranke1,SilkeSzymczak1
1InstituteofClinicalMolecularBiology,Christian‐Albrechts‐UniversityofKiel,Kiel,Germany
DNAmethylationasthebeststudiedmechanismofepigeneticmodificationofthegenomeplaysan
importantroleingeneexpression,embryonicdevelopmentanddiseasecontrol.Nowadays,next
generationsequencingtechnologiescangeneratemethylationdataforseveralmillionsofCpGsites
throughoutthegenomethatmightbespatiallycorrelated.Identifyingsinglesitesorgenomic
regionsthatenableclassificationofindividuals,e.g.ascasesorcontrolsischallenging.Weproposea
two‐steprandomforestprobabilitymachine(RFPM)approachtoselectimportantregionsandsites
withintheseregions.First,aRFPMistrainedonsitesineachregionseparately.Theestimated
probabilitybasedonthisregion(syntheticfeature)isthenusedasinputforagenome‐wideRFPM
andimportantregionsandsiteswithintheseregionsareidentifiedusingappropriatevariable
importancemeasures.WeevaluateourapproachbasedonmethylationdatasetsfromGene
ExpressionOmnibus(GEO)andTheCancerGenomeAtlas(TCGA)andcompareittoamoretime
consumingapproachusingallsitesorsummarizedmethylationratiosperregion.
P162
Statisticalapproachesforgene‐basedanalysis:Acomprehensive
comparisonusingMonte‐CarloSimulations
CarmenDering1,InkeRKönig1,AndreasZiegler1
1InstitutfürMedizinischeBiometrieundStatistik,UnversitätzuLübeck
Inrecentyearsseveralstudiesdetectedassociationsbetweengroupsofrarevariantsandcommon
diseases.Thesefindingsresultedinthedevelopmentofthe”rarevariant‐commondisease”(RVCD)
hypothesis,statingthatmultiplerarevariantstogethermaybecausalforacommondisease.
Therefore,manystatisticaltests,thecollapsingmethods,weredevelopedwhicharethetopicofthis
work.Wecomparedfifteenstatisticalapproachesinagene‐basedanalysisofsimulatedcase‐control
dataoftheGeneticAnalysisWorkshop(GAW)17invariouscollapsingscenariosand200replicates.
Scenariosdifferedinminorallelefrequency(MAF)thresholdandfunctionalityofcorresponding
collapsedrarevariants.Almostalloftheinvestigatedapproachesshowedanincreasedtype‐I‐error.
Furthermore,noneofthestatisticaltestswasabletodetecttrueassociationsoverasubstantial
proportionofreplicatesinthesimulateddata.Irrespectiveofthestatistictestused,collapsing
methodsseemtobegenerallyuselessinsmallcase‐controlstudies.Recentworkindicatesthatlarge
samplesizesandasubstantialproportionofcausingrarevariantsinthegene‐basedanalysiscan
yieldgreaterpower.However,manyoftheinvestigatedapproachesusepermutationwhichmeans
highcomputationalcost,especiallywhenapplyingagenome‐widesignificancelevel.Overcoming
theissueoflowpowerinsmallcase‐controlstudiesisachallengingtaskforthenearfuture.
P163
ApolipoproteinEgenepolymorphismandleftventricularfailureinbeta‐
thalassemia:Ameta‐analysis
NikiDimou1,KaterinaPantavou1,PantelisBagos1
1UniversityofThessaly
Thebeta‐thalassemiasyndromesareaheterogeneousgroupofgeneticdisorderscharacterizedby
reducedorabsentexpressionofthebeta‐globingene.Despiteappropriatetransfusionandchelation
therapyandlowferritinlevels,patientsstilldeveloporganfailure,heartfailurebeingthemain
causeofdeath.ApoEactsasascavengeroffreeradicals;ironchelationisprobablyanother
mechanismofitsantioxidantactivity.Thisstudywasperformedtodeterminewhetherthe
decreasedantioxidantactivityoftheapolipoproteinE(APOE)4allelecouldrepresentageneticrisk
factorforthedevelopmentofleftventricularfailure(LVF)inbeta‐thalassemiahomozygotesundera
multivariatemeta‐analysisapproach.Weincluded4studieswith613thalassemicpatientsand664
controls.Accordingtotheechocardiographicfindings,patientsweredividedintothreegroups:i)
asymptoticpatients;ii)patientswithevidenceofLVdilatation;andiii)patientswithclinicaland
echocardiographicfindingsofLVfailure.Thisclassificationschemewiththeexistenceofmultiple
groupsaswellasmultiplealleles,createdamultivariateresponseandsubsequently,theneedto
resorttomultivariatemethodsofmeta‐analysis.Wecameupwithoverallsignificantresults
contrastingE4andE3vs.E2alleleforeachgroup(Waldtest=17.14;p‐value=0.009).Multivariate
methodssuggestasignificantroleplayedbytheE4allelewhencontrastingE4allelevs.others(OR
=2.49,95%CI:1.28,4.86andOR=3.43,95%CI:1.84,6.41forgroupIIandIIIrespectively,Wald
test=16.80;p‐value<0.001).Meta‐regressionanalysisfailedtoprovideevidencethattherisk
conferredbyE4alleleisassociatedwithclinicalorhaematologicalparameters.
P164
TheCooperativeHealthResearchinSouthTyrol(CHRIS)study
CristianPattaro1,MartinGögele1,DeborahMascalzoni1,AlessandroDeGrandi1,Christine
Schwienbacher1,FabiolaDelGrecoM1,RobertoMelotti1,MaurizioFFacheris2,PeterPPramstaller1
1CenterforBiomedicine,EuropeanAcademyofBolzano(EURAC)(affiliatedInstituteoftheUniversityof
Lübeck),Bolzano,Italy
2TheMichaelJ.FoxFoundationforParkinson'sResearch,NewYork,NewYork,USA
TheCooperativeHealthResearchinSouthTyrol(CHRIS,www.christudy.it)isapopulation‐based
studytoinvestigatethegeneticetiologyofcardiovascular,metabolicandneurologicaldiseases,
startedin2011intheVenostavalley(Italy).Thepopulationischaracterizedbylong‐termsocial
stabilitywithoutmajorimmigrationevents,familiesallconnectedbyfewverylargepedigrees,and
homogeneousenvironmentalconditions.Throughacommunity‐basedcommunicationstrategy
followedbypersonalinvitation,all28,000residentadultsarebeingcontacted.Weexpectmorethan
10,000tobevoluntarilyenrolled.Eighteenself‐andinterviewer‐administeredinternationally
validatedquestionnairesreconstructtheirmedicalhistory.Electronicinstrumentalrecordings
assessfatintake,cardiacfunction,andtremor.Toenhancepowerofgene‐environmentinteraction
analyses,life‐styleexposures(nutrientintake,physicalactivity,lifecoursesmoking)areassessed
quantitatively.Urineandbloodarecollectedtomeasure19and54parameters,respectively,andfor
biobanking(cryo‐preservedurine,DNA,wholeandfractionedblood).Allparticipantswillbe
genotypedonadenseSNParray.Asubsetwillundergowhole‐genomesequencingtoidentifyrare
variantsenrichedinthispopulation.InvolvedintheP3G,BBMRI,andBioSHaREinitiatives,the
CHRISstudyandbiobankconstituteavaluableresourceforscientistswillingtoinvestigategenetic
factorsinhibitingadisease‐freeandhealthyaging.
IndexbyCategories(NeelandWilliamsAwardCandidates,Contributed
PlatformPresentationsandPosters)
Ascertainment
A2,C12,C18,C21,P104,P105
Association:CandidateGenes
C1,C3,P3,P4,P5,P6,P7,P8,P9,P10,P11,P12,P13,P14,P
15,P16,P17,P18,P19,P20,P21,P104,P105
Association:Family‐based
A2,A3,A6,C11,C22,P9,P16,P18,P22,P23,P24,P25,P26,P
27,P28,P29,P104,P105
Association:Genome‐wide
A2,A4,C1,C3,C4,C6,C7,C8,C13,C14,C16,C17,C22,P3,P
6,P13,P22,P24,P25,P27,P29,P30,P31,P32,P33,P34,P35,
P36,P37,P38,P39,P40,P41,P42,P43,P44,P45,P46,P47,P
48,P49,P50,P51,P52,P53,P54,P55,P56,P57,P58,P59,P
60,P61,P62,P63,P64,P65,P66,P67,P68,P69,P70,P71,P
72,P73,P74,P75,P76,P77,P78,P79,P80,P81,P82,P83,P
84,P85,P86,P87,P88,P89,P90,P91,P92,P93,P94,P95,P
96
Association:UnrelatedCases‐Controls
C3,C6,C9,C17,P3,P10,P16,P18,P22,P35,P43,P45,P50,P
51,P62,P64,P66,P68,P69,P70,P92,P98,P99,P100,P101,P
102,P103,P105
BayesianAnalysis
C20,P10,P20,P27,P57,P104
Bioinformatics
C3,C13,C22,P5,P25,P35,P44,P56,P60,P63,P70,P79,P
80,P86,P106,P107,P108,P109,P110,P111,P112,P113,P
114
Cancer
C9,P4,P14,P21,P30,P44,P54,P68,P72,P78,P82,P88,P
94,P101,P102,P104,P112,P115,P116,P117,P118,P119,P
120,P121
CardiovascularDiseaseand
Hypertension
A2,A4,C1,P11,P39,P49,P52,P59,P114,P122,P123,P124
Case‐ControlStudies
C2,C3,C6,C21,P3,P8,P10,P13,P21,P25,P37,P43,P45,P
54,P69,P70,P74,P82,P85,P92,P94,P98,P99,P103,P110,P
111,P112,P125,P126
Causation
C6,P25,P30,P 59,P92,P123,P127,P128
CoalescentTheory
A1,C15
CopyNumberVariation
P25,P42,P131
DataIntegration
C21,P5,P25,P30,P36,P38,P74,P79,P88,P110,P132
DataMining
C13,P25,P56,P60,P63,P80,P99,P108,P116
DataQuality
C13,C16,P109
Diabetes
A6,P56,P113,P133
EpigeneticData
C10,C21,P26,P39,P45,P79,P106,P121,P134
Epigenetics
C3,C10,C21,P5,P19,P22,P39,P45,P79,P106,P121,P134,
P135,P136
FamilialAggregationandSegregation
Analysis
P104
FineMapping
P4,P6,P25,P47,P101,P137
Gene‐EnvironmentInteraction
C22,P4,P20,P23,P42,P49,P53,P82,P102,P111,P113,P
115,P118,P131,P139,P140
Gene‐GeneInteraction
C22, P13,P16,P27,P35,P45,P46,P56,P60,P63,P70,P74,P
78,P83,P86,P107,P119,P126,P132,P138
GeneExpressionArrays
C1,C21,P71,P132
GeneExpressionPatterns
C1,C8,C10,P31,P71,P112,P132,P141
GeneticDataforClinicalTrialDesign
P139,P142
GenomicVariation
C10,C12,P8,P17,P25,P27,P30,P34,P57,P58,P68,P69,P
86,P107,P108,P109,P116,P126,P143
HaplotypeAnalysis
A6,P9,P34,P65,P114,P144
Heritability
A2,A4,C 2,C7,P84,P94,P124,P145
Heterogeneity
A1,C11,P53,P146
Homogeneity
A1,C11,P53,P146
Inbreeding
P24,P147
IsolatePopulations
P24,P147
LinkageAnalysis
A5,C11,P137,P148,P149,P150,P151,P152
LinkageandAssociation
A1,C12,P3,P9,P25,P26,P29,P32,P104,P137,P149
MachineLearningTools
P60,P63,P80,P99,P113
MarkovChainMonteCarloMethods
C20,P20,P27,P57,P104,P137
MaximumLikelihoodMethods
A4,C2,C21,P6,P29,P37,P40,P125,P127,P137
MendelianRandomisation
C9,P59,P122,P123,P128
MicrobiomeData
C19,C20,P153
MissingData
C16,P32,P57
C7,P23,P24,P26,P68,P102,P126
MultilocusAnalysis
C2,C4,P3,P27,P57,P73,P77,P78,P81,P98,P119,P126,P
137
MultipleMarkerDisequilibrium
Analysis
A3,P3,P41,P58,P144
MultivariatePhenotypes
C3,C20,P15,P50,P55,P76,P84,P90,P96,P98,P126,P127,
P136
Pathways
C2,P15,P27,P35,P54,P56,P68,P71,P78,P83,P119
PopulationGenetics
C3,C18,P13,P27,P43,P56,P57,P94,P125,P126,P146,P
147,P151,P154,P155
PopulationStratification
C3,C12,C17,P33,P90,P146,P154,P155,P156
PredictionModelling
C7,C8,P31,P34,P56,P73,P90,P125,P142,P157
PsychiatricDiseases
C6,P19,P41,P87,P90,P113,P126,P128,P158
QuantitativeTraitAnalysis
A2,A6,C5,C6,C10,C14,C19,C20,P6,P12,P22,P27,P37,P
50,P52,P57,P65,P68,P71,P73,P84,P86,P91,P119,P126,P
127,P159
SampleSizeandPower
P29,P43,P53,P125,P159
SequencingData
A2,A3,A5,C11,C12,C13,C14,C16,P3,P8,P9,P25,P27,P
28,P57,P70,P79,P98,P100,P101,P108,P109,P112,P116,P
150,P151,P155,P158
TransmissionandImprinting
P22,P26,P87
IndexbyAuthors
Abel,HaleyJ
Abel,Laurent
Adeyemo,Adebowale
Agarwal,Anita
Ahmad,Wasim
Allanore,Yannick
Amos,ChristopherI
Amouyel,Philippe
Anand,SoniaS
Andelfinger,Gregor
A1
P151
C18,P77,P81
C2
C11
P83
E4,C9,P4,P44,P54,
P78,P101,P119,P
112
P24
P122
P147
Anderson,Carl
Andreassen,BettinaK
Andrew,AngelineS
Antounians,Lina
Aquino‐Michaels,Keston
Arbeev,KonstantinG
Arbeeva,LiubovS
Arbet,Jaron
P67
P38
P121
P79
C8
P48,P93
P48,P93
P69
Armasu,SebastianM
Asimit,JenniferL
Asselbergs,Folkert
Auer,PaulL
Avinash,Avinash
Avril,Marie‐Françoise
B,Poonkuzhali
Babron,Marie‐Claude
P33
P61
P86
P36
P151
P78,P119
P126
P24
Bagos,Pantelis
Bailey‐Wilson,JoanE
P89,P163
M2,P3,P60,P108,P
116
P66
P14,P102
C21
P158
P53
Baksh,MFazil
Balavarca,Yesilda
Balliu,Brunilda
Bansal,Vikas
Barata,Llilda
Barrdahl,Myrto
Barrett,JenniferH
Barrington‐Trimis,Jessica
L
Barroso,Inés
Bartlett,ChristopherW
P117,P118
C9
P82
P61
P137
Barton,SheilaJ
Bayas,Antonios
Beaty,TerriH
Becker,Natalia
Becker,Tim
Beilby,John
Belkadi,Aziz
Below,JenniferE
P23
P75
P29
P14
P46,P74
P124
P151
P149
Bendlin,BarbaraB
Benitez,Alejandra
Benner,Christian
Bentley,Amy
Berg,Richard
Bergametti,Françoise
Bergen,Andrew
Bergh,FlorianT
P113
P69
A4
P77,P81
P111
P150
P158
P75
Bergsma,WicherP
Bernatsky,Sasha
Berndt,Sonja
Berrettini,Wade
Bessonov,Kyrylo
Bettecken,Thomas
Bickeböller,Heike
Bielinski,Sue
P125
P136
P61
P158
P5,P132
P75
P54
C4
Biemans,Floor
Biernacka,JoannaM
Bijma,Piter
Bishop,DavidT
Blangero,John
Blue,Elizabeth
Bock,Christoph
Boehringer,Stefan
P34
P99
P34
C9
P124
C12
I3
C21
Boisson,Bertrand
Borecki,Ingrid
Bouzigon,Emmanuelle
Bradford,Yuki
Brand,Bodo
Brantley,MilamA
Brennan,Paul
Brenner,Hermann
P151
P53
P26
C4
P71
C2
P4,P54,P101
P14
Brilliant,Murray
Brossard,Myriam
Brugger,Markus
P35,P111
P23,P78,P119
P152
Chuang,Lee‐Ming
P13
Ciampi,Antonio
P136
Clerget‐Darpoux,Françoise C7
Buck,Dorothea
Buck,Katharina
Bull,Shelley
Burt,Amber
Burwinkel,Barbara
Bush,WilliamS
Butterbach,Katja
C,Chithra
P75
P102
P40,P127
P33
P14,P102
C2,C4
P102
P126
Cleves,MarioA
Coassin,Stefan
Colmegna,Ines
Connolly,Siobhan
Cook,JamesP
Cooney,KathleenA
Cooper,DavidN
Cooper,RichardS
P16,P18
P12
P136
P87
P43,P52
P116
P64
P11
Cadby,Gemma
Calhoun,Vince
Campa,Daniele
Candille,SophieI
Canzian,Federico
Cao,Guichan
Cao,Hongbao
Capanu,Marinela
P124
P110
P117
C17
P117
P11
P110
P8
Coram,Marc
Cordell,HeatherJ
Cortessis,VictoriaK
Cowen,Philip
Cox,NancyJ
Crawford,AndrewA
Crawford,DanaC
Croitoru,Kenneth
C17
P1,P67,P148
P104
P95
C8
P95
C4
C19
Carland,Tristan
Carpten,JohnD
Carter,ToniaC
Casanova,Jean‐Laurent
Casas,Juan‐Pablo
Chaichoompu,
Kridsadakorn
Chalise,Prabhakar
P158
P116
P3
P151
P123
P154
P116
C4
P48
P68
P49
P49
P94
Chan,Andrew
Chan,JohnnyChun‐Yin
Chang‐Claude,Jenny
Chareanchim,Wanwisa
Charoen,Pimphen
Chen,BiLing
Chen,Guanjie
Chen,WeiV
P75
P120
P14,P118
P154
P123
P135
P77,P81
P119
Cropp,CherylD
Crosslin,DavidR
Culminskaya,Irina
Cunningham,JulieM
Cupples,L.Adrienne
Czajkowski,Jacek
daSilvaFilho,Miguel
Inacio
Daar,EricS
Dai,Hang
Daley,Denise
Danecek,Petr
Darabos,Christian
Darst,BurcuF
Davey‐Smith,George
deAndrade,Mariza
Chen,Zhijian
Chernikova,DianaA
Cheung,BernardMan‐
Yung
Cheung,Ching‐Lung
Chien,Li‐Chu
Chiu,Yen‐Feng
Christensen,Kaare
P127
P153
P120
Christiani,DavidC
Chu,Shih‐Kai
P4
P65
deGrandi,Alessandro
deJong,MartMC
deLima,RenataLLF
deSilva,Niletthi
De,Rishika
Dedoussis,George
Degenhardt,FraukeC
D'Elia,Yuri
P145,P164
P34
P2
P59
P86
C13
P161
P145
Demearath,EllenW
P39
P88
P120
P13
P13
P58
C3
A5
P135
C13
P56
P113
P22
P33
Feng,ZenyZ
Fernandes,Elisabete
Fernández‐Rhodes,
Lindsay
Field,JohnK
Field,L.Leigh
Filipits,Martin
Fischer,Christine
Fischer,Krista
Fisher,Virginia
Flaquer,Antonia
P4,P101
P108
M1
C16
I2,C10,P133
P49
C6
Fleischer,Sabine
Foraita,Ronja
Fornage,Myriam
Forrester,Terrence
Försti,Asta
Fouladi,Ramouna
Franceschini,Nora
Francis,Ben
P75
P140,P141
P39
P11
P94
P5,P154
P47
P90
Franke,Andre
Franke,Lude
Fraser,Abigail
Freitag‐Wolf,Sandra
Fridley,BrookeL
Friedman,ThomasB
Frigessi,Arnoldo
Froguel,Philippe
P75,P161
C10
P59
P156
E1,P88
C11
P38
P159
Fu,Wenjiang
Fuchs,Michael
G,Anilkumar
Gadaleta,Francesco
Gagné‐Ouellet,Valérie
Gagnon,France
Gamazon,EricR
Garcia‐Closas,Montserrat
P70
P94
P19
P132
P23
P79
C8
P118
P128
P164
P3
P78,P119
C13
P53,P58
Gaunt,Thomas
Gauvin,Héloïse
Gazal,Steven
Geng,Ziqian
Génin,Emmanuelle
Ghasemi‐Dehkordi,Payam
Ghosh,Saurabh
Gieger,Christian
P59
P147
P24,P150
I4
C7,P24
P17
P7,P50
P12
P55
Gilbert‐Diamond,Diane
P86
Demenais,Florence
Deng,Bo
Deng,Xuan
C9,P78,P119
P68
P49
Dennis,Jessica
Denroche,RobertE
Dering,Carmen
Desch,KarlC
Dimou,Niki
Dizier,Marie‐Hélène
D'Mello,Matthew
Doheny,Kimberly
P79
P101
P162
P84
P89,P163
P23,P26
P122
P25,P109
dosAnjosSilva,LuziaP
Doumatey,Ayo
Drenos,Fotios
Drichel,Dmitriy
Drummond,MeghanC
Duan,Qing
Dudbridge,Frank
Dudek,ScottM
P2
P77,P81
P52,P86
P74
C11
P39
P37,P72,P123
P30
Dupuis,Josée
Durazo‐Arvizu,Ramon
Easton,Douglas
Eichler,EvanE
Eikelboom,John
Ekstrøm,ClausT
Elhezzani,NajlaS
Ellis,George
A6
P11
P72,P118
P28
P122
C22
P125
P135
Emeny,Rebecca
Engelman,CorinneD
Engert,Andreas
Erdmann,Jeanette
Erickson,StephenW
Esko,Tõnu
Esparza‐Gordillo,Jorge
Espin‐Garcia,Osvaldo
C6
P113
P94
P51
P16,P18
C10,P133
P26
P127
Eu‐ahsunthornwattana,
Jakris
Evans,Jonathan
Facheris,Maurizio
Fan,Ruzong
Fang,Shenying
Farmaki,Aliki‐Eleni
Feitosa,MaryF
P148
Feng,Tao
P96
P6
P39
Gilly,ArthurL
Ginsburg,David
Girirajan,Santhosh
C13
P84
P42,P131
Haun,Margot
Hayes,GeoffreyM
Hayward,NicholasK
P12
P33
C9
GirondoRodriguez,Mar
Gögele,Martin
Gold,Ralf
Goldberg,Jack
Gomez,Felicia
Goodarzi,MarkO
Göpel,Wolfgang
Gorlov,IvanP
A2
P145,P164
P75
P134
P22
A6
P62
P4,P44
He,Chunyan
He,Liang
He,Zong‐Xiao
Heard‐Costa,Nancy
Heesen,Christoph
Heibati,Fatemeh
Heid,IrisM
Heit,JohnA
P76
P27,P57
P28
P49
P75
P17
P53
P33
Gorlova,OlgaY
Gosset,Simon
Goulet,JosephL
Graff,Mariaelisa
Grallert,Harald
Grätz,Christiane
Greco,Brian
Greene,CaseyS
P44
P150
P41
P39,P49,P53
C1,C6
P75
P69
P63
Hemmer,Bernhard
Hemminki,Kari
Herder,Christian
Hermann,BruceP
Herms,Stefan
Herold,Christine
Heron,ElizabethA
Herting,Egbert
P75
P94
C1
P113
P94
P46,P74
P87,P92
P62
Greenwood,CeliaMT
Grinde,Kelsey
Grove,MeganL
Guan,Shunjie
Guan,Weihua
Guey,Stephanie
Gulick,RoyM
Günther,Frauke
P136
P69
P39
P70
P39
P150
C3
P140
Hertz‐Picciotto,Irva
Heßler,Nicole
Hetmanski,JacquelineB
Hetrick,Kurt
Hibberd,PatriciaL
Hill,Doug
Hirschfield,GideonM
Hirschhorn,JoelN
P42,P131
P160
P108
P25,P109
P153
P63
P67
P22
Guo,Wei
Gusareva,ElenaS
Gwozdz,Wencke
Habermann,Nina
Haessler,Jeffrey
Haines,JonathanL
Hainline,Allison
Hall,IanP
P142
P45
P140
P102
P47
C2,P35
P69
P43,P91
Ho,Ing‐Kang
Hobbs,CharlotteA
Hodgson,Karen
Hoen,AnneG
Hoffmeister,Michael
Hofmann,Per
Hoggart,CliveJ
Holmes,Michael
P65
P16,P18
P95
P153
P14,P102
P94
P22
P86
Hall,JacobB
Hall,MollyA
Hall,Per
Hamann,Ute
Hamon,Julie
Hashemzadeh‐Chaleshtori,
Morteza
Hass,DavidW
C2
P42,P45,P111,P131
P118
C15,C16
P83
P17
P75
P60,P108
C1
P114
P153
A2,C21,P105
C3
Holste,Theresa
Holzinger,EmilyR
Homuth,Georg
Hosseini,BitaSadat
Housman,MollyL
Houwing‐Duistermaat,
Jeanine
Howey,Richard
Haubrich,Richard
C3
Hsu,Yi‐Hsiang
P76
P1
Hu,Yijuan
Huang,Jie
Hudson,Marie
P36
C14
P136
Kilpeläinen,Tuomas
Kim,DanielS
Kim,Dokyoon
P49
P33
P30,P42,P131
Huff,ChadD
Hui,Jennie
Hung,Joseph
Hung,RayjeanJ
Hwang,Heungsun
Igl,WilmarM
Iles,MarkM
Im,HaeKyung
P149
P124
P124
P4,P54,P101
P15
P10
C9
E3,C8
Kirdwichai,Pianpool
Kirubakaran,Richard
Kitchner,Terrie
Kloss‐Brandstätter,Anita
Koestler,DevinC
König,InkeR
Koscik,RebeccaL
Koslovsky,MatthewD
P66
P21
P111
P12
P153
P51,P162
P113
P20
Ionita‐Laza,Iuliana
Itan,Yuval
Jackson,LatifaF
Jackson,VictoriaE
Jacob,KuruthukulangaraS
Jacob,Molly
Jäger,MartinL
James,AlanL
P8,P100
P151
P107
P91
P126
P126
P141
P124
Kovach,JaclynL
Kramer,Holly
Krawczak,Michael
Kieseier,BerndC
Kronenberg,Florian
Krumm,Niklas
Krystal,JohnH
Kuivaniemi,Helena
C2
P11
P64,P156
P75
P12
P28
P41
C4,P33
Jansen,Lina
Jeanpierre,Marc
Jenkins,Gregory
Jiang,Congqing
Johnson,Michael
Johnson,SterlingC
Jomphe,Michèle
Jones,Dean
P14
P144
P99
P77
P90
P113
P147
P134
Kulkarni,Hemant
Kullo,Iftikhar
Kulminski,AlexanderM
Kung,AnnieWai‐Chee
Kuo,Hsiang‐Wei
Kuruvilla,Anju
Kutalik,Zoltán
Kümpfel,Tania
P7
C4
P48,P93
P120
P65
P126
P22,P32,P53
P75
Jorgensen,Andrea
Jöckel,Karl‐Heinz
Just,Jocelyne
Justice,AmyC
Justice,Anne
Justice,CristinaM
Kaaks,Rudolf
Kabisch,Maria
P90
P75
P23
P41
P49,P53
P143
P117
C15,C16
Kwon,Soonil
Labbe,Aurélie
Labuda,Damian
Lacour,Andre
Ladwig,Karl‐Heinz
Laisk‐Podar,Triin
Läll,Kristi
Lambert,Jean‐Charles
C1
P15,P136
P147
P74
C6
P103
P133
P24
Kachuri,Linda
Kap,ElisabethJ
Karaderi,T
Karyadi,Danielle
Kasela,Silva
Kaye,Walter
Keane,Thomas
Keating,BrendanJ
P101
P14,P102
P103
P116
C10
P158
C14
P45,P86
Lambourne,Stephan
BuscheJohn
Lamina,Claudia
Laprise,Catherine
Lathrop,Mark
Lauria,Fabio
Lavielle,Nolwenn
Law,MatthewH
P136
P12
P26
P23,P78
P140
P23
C9
Kedenko,Lyudmyla
P12
Lawlor,Debbie
P59
Leal,SuzanneM
Leavy,Olivia
LeBlanc,Marissa
A5,C11,P28,P155
P72
P38
Luke,Amy
M,FabiolaDelGreco
Ma,Qianyi
P11
P164
P84
Lee,JeffreyE
Lefebvre,Jean‐François
Legrand,Carine
Leiro,MarcelaMQL
LeMarchand,Loïc
Leslie,ElizabethJ
Leutenegger,Anne‐Louise
Lewis,Glyn
P78,P119
P147
P146
P2
P4,P101
P9
P24
P95
MacLeod,StewartL
Madan,JulietteC
Mägi,Reedik
P18
P153
P85,P103,P106,P
133
P158
P47,P85
P50
P26
Lewis,Sarah
Li,Biao
Li,Jing
Li,Jingyun
Li,JunZ
Li,Ming
Li,Qing
Li,Ruowang
P95
A5
P121
P16,P18
P84
P16
P108
P30
Li,ShuweiS
Li,Yafang
Li,Yun
Liang,Kung‐Yee
Lieb,Wolfgang
Lin,Danyu
Lindgren,CeciliaM
Lindström,Sara
P155
P4,P101,P112
P36,P39
P13
P75,P156
P36
P103
P117
Ling,Hua
Linneman,Jim
Lishout,FrancoisVan
Liu,Aiyi
Liu,Geoffery
Liu,Guozheng
Liu,Sheng‐Wen
Liu,Yu‐Li
P25,P109
P35
P5
C5
P101
P77
P65
P65
Lobach,Iryna
Lokk,Kaie
Loley,Christina
Long,Quan
Loos,RuthJF
LorenzoBermejo,Justo
Lu,Qing
Lucas,AnastasiaM
P3
P106
P51
C5
P22,P53
C15,C16,P146
P98
P33,P35
Lui,VivianWai‐Yan
P120
Magistretti,Pierre
Mahajan,Anubha
Majumdar,Arunabha
Malerba,Giovanni
Malik,Sadia
Malley,James
Mangino,Massimo
Mangold,Elisabeth
Marazita,MaryL
Marchini,Jonathan
Margaritte‐Jeannin,
Patricia
Marson,Anthony
MARTINEZ,Maria
Masca,NicholasGD
Mascalzoni,Deborah
Matchan,Angela
Matteini,AmyM
Maubec,Eve
Mayeux,Richard
P18
P60
P22
P108
P108
P43
P23
P90
P83
P52
P164
C13
P58
P119
P58
McCallum,Kenneth
McCarthy,Shane
McCarty,CatherineA
McDonnell,ShannonK
McGuffin,Peter
McKay,JamesD
McKenzie,ColinA
McPhersonJonn
Medina‐Rivera,Alejandra
Meisinger,Christa
Meitinger,Thomas
Mells,George
Melotti,Roberto
Melton,PhillipE
Memari,Yasin
Merikangas,AlisonK
P100
C14
P35,P45,P111
P116
P95
P4,P101
P11
P101
P79
C6
C6
P67
P164
P124
C14
P92
Metspalu,Andres
Middha,Sumit
C10,P103,P106
P116
Mihailov,Evelin
Milani,Lili
Mills,JamesL
P52
C10
P3
Milne,Roger
Min,Josine
Minster,RyanL
Moffatt,Miriam
Mohamdi,Hamida
Moore,JasonH
P118
C14
P58
P26
P78,P119
P5,P44,P45,P56,P
63,P121,P153
P17
Moradipour,Negar
Morange,Pierre
Moreau,Claudia
Morin,Andréanne
Morris,AndrewP
Panoutsopoulou,Kalliope
Pantavou,Katerina
Parchami‐Barjui,
Shahrbanuo
Pare,Guillaume
Parker,MargaretM
Pastinen,Tomi
Paternoster,Lavinia
Paterson,AndrewD
Pattaro,Cristian
Paul,Fiedemann
P61
P163
P17
P122
P108
P136
P128
C19,C20,P127
P145,P164
P75
Morrison,HilaryG
Moses,EricK
Müller,C
P79
P147
P26
P43,P47,P61,P85,P
90,P103
P153
P124
C1
Paulweber,Bernhard
Pedergnana,Vincent
Peil,Barbara
Peissig,Peggy
Pendergrass,SarahA
Müller,Christian
Müller‐Myhsok,Bertram
Munroe,PatriciaB
Musk,ArthurW
Na,Jie
Nadif,Rachel
Najafi,Mohammad
Neumann,Christoph
C1
M3,P75
P52
P124
P68
P23
P114
P29
Perry,John
Peters,Annette
Peters,Marie
Peters,TimJ
Peterson,Pärt
Peto,Julian
Petterson,TanyaM
Pfuetze,Katrin
C14
C6,P12
P85
P95
C10
P72
P68
P102
Newman,AnneB.
Nick,ToddG
Nickerson,DeborahA
Nitsch,Dorothea
Noce,Damia
North,KariE
Nöthen,MarkusM
Nothnagel,Michael
P58
P16
C11
P123
P145
P53
P75,P108
P64
Pharoah,PaulDP
Pigeot,Iris
Pineau,ChristianA
Pirinen,Matti
Pitkäniemi,Janne
Posch,Martin
Posevitz,Vilmos
Pramstaller,PeterP
C9,P118
P140
P136
A4
P27,P57
I1
P75
P145,P164
Ntalla,Ioanna
Nutt,David
O'Donovan,MichaelC
O'Roak,BrianJ
Oros,KathleenKlein
Orr,Nick
Ozel,AyseB
P,SujithaS
P43
P95
P95
P28
P136
P118
P84
P19
Preuss,Christoph
Preuß,Michael
Province,MichaelA
Pugh,Elizabeth
Qiu,Jingya
R,Anand
R,Veeramanikandan
Rahmioglu,Nilufer
P147
P62
A1
P25,P109
P56
P21
P21,P126
P85
Palla,Luigi
Pankow,James
P37,P72
P39
Rajkumar,AntoP
Ramirez,Alfredo
P126
P46
P12
P151
C16
P35,P111
C3,C4,P42,P45,P
86,P111,P131
Pericak‐Vance,MargaretA C2
Perls,ThomasT
P58
Rashki,Ahmad
Ray,Ajit
Rehman,AtteeqU
P17
P108
C11
Sanders,JasonL
P58
Sandford,Richard
P67
Santos‐Cortez,RegieLynP C11
Reinsch,Norbert
Reisch,LuciaA
Riazuddin,Saima
Riazuddin,Sheikh
Ripatti,Samuli
Ritchie,GrahamRS
Ritchie,MarylynD
Rivolta,Carlo
Robbins,GregoryK
Roberts,Robert
Robertson,Neil
Rocheleau,Ghislain
Rohde,Klaus
Romdhani,Hela
Romm,Jane
P71
P140
C11
C11
A4,P27,P57
C14
C3,P30,P42,P45,P
86,P111,P131
P22
C3
P122
P85
P159
P26
P15
P109
Sarin,AnttiP
Sarnowski,Chloé
Sayers,Ian
Scanlon,PaulD
Scheet,Paul
Scheinhardt,MarkusO
Scherer,Dominique
Schillert,Arne
P57
P23,P26
P91
P68
I4
P71
P102
C1
Rosenberger,Albert
Ross,Stephanie
Rotimi,CharlesN
Rousson,Valentin
Roustazadeh,Abazar
Roy‐Gagnon,Marie‐Hélène
Ruczinski,Ingo
Rudi,Knut
P54
P122
C18,P77,P81
P22
P114
P147
P29
I5
Schmidt,Marjanka
Schork,Nicholas
Schramm,Katharina
Schüller,Vitalia
Schupf,Nicole
Schurmann,Claudia
Schwantes‐An,Tae‐Hwi
Schwartz,StephenD
P118
P158
C1
P46
P58
C1
P73,P143
C2
Rudolph,Anja
Rue,AsenathLa
Rüeger,Sina
Russo,Paola
S,Gunasekaran
S,Harshavaradhan
S,Lakshmanan
S,Lakshmikirupa
P118
P113
P32
P140
P19
P19
P19
P126
Schwartzentruber,Jeremy
Schwender,Holger
Schwerin,Manfred
Schwienbacher,Christine
Scott,WilliamK
Segurado,Ricardo
Seibold,Petra
Seldin,Mike
C13
P29
P71
P145,P164
C2
P92
P14,P102
P67
Saad,Mohamad
Saare,Merli
Sabourin,JeremyA
Sager,MarkA
Sahbatou,Mourad
Sakuntabhai,Anavaj
Sallis,Hannah
Salumets,Andres
A3
P85
P143
P113
P24
P154
P128
P85,P103,P106
Selleck,ScottB
Severi,Gianluca
Sham,Pak‐Chung
She,Jun
Shen,Mengtian
Shete,Sanjay
Shih,PBetty
Shin,Ji‐Hyung
P42,P131
P4
P120
P68
P70
P138,P157
P158
P40
Samani,NileshJ
P122
Shrine,Nick
Shriner,Daniel
Shugart,YinYao
Shui,Irene
Siani,Alfonso
Siegert,Sabine
Sillanpää,MikkoJ
Sills,Graeme
P43
C18,P77,P81
P110
P117
P140
P64
P27,P57
P90
Siminovitch,KatherineA
P67
Sing,Chor‐Wing
Smith,GeorgeDavey
Sobreira,Nara
P120
P128
P25
Tissier,RenaudR
Tobin,MartinD
Todd,Nick
P105
P43,P52,P91
P18
Sogin,MitchellL
Sorant,AlexaJM
Southam,Lorraine
Spitz,Margaret
Srivastava,Alok
Srivastava,Amitabh
SS,Prakash
Stahl,EliA
P153
P73,P143
C13
P157
P126
P121
P21
C5
Tongsima,Sissades
Tõnisson,Neeme
Tornaritis,Michael
Tournier‐Lasserve,
Elisabeth
Tozeren,Aydin
Trégouët,David
Tromp,Gerard
P154
P106
P140
P150
Stallard,Eric
Stanford,Janet
Stangel,Martin
Staples,Jeffery
Steen,KristelVan
Steer,Colin
Stewart,AlexanderFR
Stewart,WilliamCL
P48
P116
P75
P149
P5,P132,P154
P128
P122
P137
Truong,Vinh
Tserel,Liina
Tsonaka,Roula
Tumani,Hayrettin
Turpin,Williams
Tworoger,Shelley
Ukraintseva,SvetlanaV
Ul‐Abideen,MuhammadZ
P79
C10
C21,P105
P75
C19
P101
P48,P93
P153
Strachan,DavidP
Strauch,Konstantin
Sun,YanV
Sun,Zhifu
Sundquist,Jan
Sung,Heejong
Swartz,MichaelD
Szymczak,Silke
P43
C6,P12,P75,P152
P134
P68
P94
P73,P143
P20
P60,P161
Ulrich,CorneliaM
Ushey,Kevin
Vaccarino,ViolaL
Vaitsiakhovich,Tatsiana
Valle,David
Vandewater,ElizabethA
Vaysse,Amaury
Veidebaum,Toomas
P14
P135
P134
P46,P74
P25
P20
P78,P119
P140
Tachmazidou,Ioanna
Tackenberg,Björn
Táernikova,Natalia
Talluri,Rajesh
Tammiste,Triin
Tan,KathrynChoon‐Beng
Tang,Hua
Tang,Xinyu
C14
P75
P106
P157
P103
P120
C17
P16,P18
Venturini,Giulia
Verdura,Edgard
Verma,Anurag
Verma,ShefaliS
Vézina,Hélène
Vincent,Quentin
Volk,Heather
P22
P150
C3,C4,P35
C4,P33,P35,P42,P
45,P86,P131
P147
P26,P151
P42,P131
Taub,MargaretA
Tayo,BamideleO
Tekola‐Ayele,Fasil
Thibodeau,StephenN
Thomas,DuncanC
Thomsen,Hauke
Thromp,Gerard
Tikkanen,Emmi
P9,P29
P11
C18
P116
P104
P94
P33
A4
vonStrandmann,ElkeP
Vrabec,TamaraR
Wagner,ErinK.
Wahl,Simone
Wain,LouiseV
Waldenberger,Melanie
Wallace,JohnR
Walter,Klaudia
P94
P35
P76
C1
P43,P52,P91
C6.P75
P45,P111
C14
Tintle,NathanL
P69
Wampfler,JasonA
Wang,GaoT
P68
A5,P28,P155
P107
P79
P35
Wang,Sheng‐Chang
Wang,Shuai
Wang,Yifan
P65
A6
P3
Wu,Deqing
Xia,Jin
Xiao,Xiangjun
P93
P76
P112
Wang,Yu‐Ping
Wang,Zhong
Wang,Zuoheng
Wangkumhang,
Pongsakorn
Warren,Helen
Weber,Frank
Weal,Mike
P110
P41
P41
P154
P52
P75
P125
Weeks,DanielE
Wei,Changshuai
Wei,Peng
Wei,Qingyi
Weidinger,Stephan
Wendt,Christine
Westra,Harm‐Jan
Wheeler,Eleanor
P3
P98
P115
P78
C6
P68
C10
P61
Xiong,Momiao
Xu,Ke
Xu,Lizhen
Xu,MengYuan
Xu,Wei
Yadav,Pankaj
Yang,Hsin‐Chou
Yao,Chen
P3
P41
C19,C20
P142
C19,C20
P156
P65
P113
Wiendl,Heinz
Wijsman,EllenM
Wilantho,Alisa
Wilch,EllenS
Wildemann,Brigitte
Wiles,Nicola
Wilkinson,AnnaV
Wilson,AlexanderF
P75
A3
P154
C11
P75
P95
P20,P157
P3,P73,P143
Yao,Yin
Yashin,AnatoliyI
Yengo,Loïc
Young,KristinL
Younkin,SamuelG
Zeeland,AshleyVan
Zeggini,Eleftheria
Zettl,Uwe
P142
P48,P93
P159
P49
P29
P158
C13,P43,P61
P75
Wilson,Michael
Winham,StaceyJ
Winkler,ThomasW
Witmer,Dane
Witte,John
Wojczynski,MaryK
Wolf,Andreas
Wong,IanChi‐Kei
P79
P99
P49,P53
P25
E2
P58
P64
P120
Zhang,Di
Zhang,Guosheng
Zhang,Peng
Zhang,Shuo
Zhang,Zhenjian
Zhao,JingH
Zhou,Jie
Zhou,Yanxun
A5,P155
P39
P25,P109
P68
P77,P81
A6
P81
P77,P81
Zhu,Jun
Zhu,Xiaofeng
Ziegler,Andreas
Wong,William
Wright,MarvinN
Wu,Chih‐Chieh
Wu,ColinO
P96
P80
P138
C5
C5
P55
M4,C1,P51,P62,P
71,P75,P80,P139,P
160,P162
P71
P75
I4
P85
P38
P101
Zimmer,Daisy
Zipp,Frauke
Zöllner,Sebastian
Zondervan,KrinaT
Zuber,Verena
Zuzarte,PhilipC

2014 IGES Scientific Meeting Abstracts

Transcription

Similar documents

here - Center for Genetics and Society

Linking Genotype to Phenotype

What has changed - Center for Genetics and Society

Barbara Schoenfeld

Dragonfly genome project

PERNIGOTTI | Savor a taste of Italy with Pernigotti Gelato

Presentation - Produce Marketing Association

Exagen Genomic Technology

DNA sequencing and