Bayesian multi-phenotype genome-wide association for structured experimental designs. Anthony J. Greenberg^1,2, Gabriel E. Hoffman¹, Pavel Korniliev¹, Yuxin Shi¹, Susan McCouch², Jean-Luc Jannink², Jason Mezey¹. 1) Dept BSCB, Cornell Univ, Ithaca, NY; 2) Dept PBG, Cornell Univ, Ithaca, NY.

   The quest to understand the relationship between genetic and phenotypic variation underpins most fundamental research in evolutionary genetics. It is also critically important for progress in applied fields, such as plant and animal breeding to improve efficiency of food production. With the development of dense single nucleotide polymorphism (SNP) markers, it has become possible to assess genotype-phenotype relationships by conducting genome-wide association (GWA) studies. However, these studies are statistically and computationally challenging due to the presence of confounding factors, such as population structure, and the massive amount of SNP and phenotype data. In model organisms, such as Drosophila, phenotypic measurements for GWA are typically performed on inbred lines that are reared in a replicated experiment. Statistical methods developed so far are typically unable to both account for confounding effects and deal for complex experimental designs, especially when multiple phenotypes are considered at once. We developed a Bayesian hierarchical approach that jointly estimates quantitative-genetic parameters for multiple phenotypes, corrects the influence of population structure, and estimates pleiotropic and single-phenotype effects of individual SNPs. The method can incorporate experimental designs of arbitrary complexity, including crosses among lines. We test our method on extensive simulations and apply it to a data set of wing measurements in a set of D. melanogaster lines from the DGRP collection.