Poster 44
Presenter: Jeremy Sabourin
Thursday, 4:00 – 6:00pm
Jeremy Sabourin and William Valdar University of North Carolina at Chapel Hill
Highly recombinant populations derived from multiple founder crosses, such as heterogeneous stocks, can be used to map loci far more accurately than possible with standard intercrosses. However, the varying degree of relatedness that exists between individuals complicates the analysis. Incorporating polygenic effects based on known or inferred kinship can provide an efficient solution for significance testing at single loci. But several groups have shown that this can be improved in several ways: by using kinship informed by previous model selection, and by explicitly modeling multiple loci. Moreover, the recent consensus in the animal breeding community, and elsewhere, suggests explicit model selection even in the absence of kinship-like modeling may be beneficial. We add to this literature by exploring the use of model selection and model averaging techniques applied over the entire genome, comparing these with some popular alternatives. In our method, LLARRMA-haplo, we select multiple haplotype locus association models using a least absolute shrinkage and selection operator (LASSO) based regression applied to resampled data sets. We provide resample model averaging statistics about the probability of haplotype regions being included under model selection, providing a highly generalizable frequentist alternative to Bayesian model selection This often leads to more accurate identification of haplotype regions than by single-locus mapping with polygenic effects. The generality of our approach means it can potentially be applied to any multiple founder population of unknown structure.