Continuous models for mixtures
Transcription
Continuous models for mixtures
Continuous models for mixtures Thore Egeland Copenhagen April 20-23 2015 Continuous models Norwegian University of Life Sciences 1 http://www.cstl.nist.gov/strbase/training/ISFG2013workshops.htm Continuous models Norwegian University of Life Sciences 2 Continuous models Norwegian University of Life Sciences 3 Binary model 1. Qualitative binary model (aka unrestricted…) –Treats alleles as present or absent and does not take into account peak height information 2. Semi-quantitative binary model (aka restricted …) –Declares some of the combinations as possible or impossible Continuous models Norwegian University of Life Sciences 4 Example 6 genotype combinations {7/9,11/13},{7/11,9/13},{7/13,9/11} {11/13,7/9},{9/13,7/11},{9/11,7/13} Continuous models Norwegian University of Life Sciences 5 Suspect 7/11 H P : S+NN1, H D : NN1 + NN2 Assume only combinations are 7/11, 9/13 2 p9 p13 1 LR2 = 2 p7 p9 2 p11 p13 + 2 p11 p13 2 p7 p9 4 p7 p11 Ignoring peak information 2 p9 p13 1 = LR1 = 24 p7 p9 p11 p13 12 p7 p11 LR2 = 3LR1 Continuous models Norwegian University of Life Sciences 6 Semi continuous models • Artefacts: –Drop-out –Drop-in –Stutters Continuous models Norwegian University of Life Sciences 7 Semi-continuous model include a drop-out probability d Software: LRmixStudio (NFI, Haned,...), .... Continuous models Norwegian University of Life Sciences 8 Continuous model 6 genotype combinations {7/9,11/13},{7/11,9/13},{7/13,9/11} {11/13,7/9},{9/13,7/11},{9/11,7/13} weighed according to a probability distribution Continuous models Norwegian University of Life Sciences 9 Suspect 7/11 H P : S+NN1, H D : NN1 + NN2 w2 2 p9 p13 LR = w1 2 p7 p9 2 p11 p13 + w2 2 p7 p11 2 p9 p13 + ... + w6 2 p9 p11 2 p7 p13 Qualitative binary model: w1= ...= w6= 1 w3 = w4 = w6 = Semi − quantitative binary model: w2 = 1, w6 = 1, w1 = 0 Continuous models Norwegian University of Life Sciences 10 Computation LR = w2 2 p9 p13 w1 2 p7 p9 2 p11 p13 + ... + w6 2 p9 p11 2 p7 p13 + drop − terms • Hard part: Modelling and estimating weights w j = Pr(peak heights|known contributors) Bayesian networks (Graversen, Cowell, Lauritzen, ... o DNAmixtures MCMC (Taylor, Bright,...) o STRMIX http://strmix.esr.cri.nz/ gammadnamix efm. http://rpackages.ianhowson.com/rforge/gammadnamix/ (Bleka) Continuous models Norwegian University of Life Sciences 11 © ESR 2013