Project summary Understanding the genetic architecture of complex traits is critical for personalized medicine, as knowledge of underlying risk loci is required for treatment and prediction of disease state. Genome wide association studies (GWAS) have identified hundreds of thousands of genotype-phenotype associations. This information – incorporated in Polygenic Risk Scores (PRS) – has provided levels of prediction accuracy for some complex diseases and health-related traits comparable to those obtained for monogenic diseases. While this is an important advance, significant problems still need to be overcome to achieve the ultimate goal of precision medicine. First, prediction accuracy is low for most complex traits. Possible reasons for this observation are the higher polygenicity of these traits, requiring larger sample sizes to discover associated loci and estimate their effect; and gene-by-context interactions that are usually ignored in prediction models. Second, genetic analyses of complex traits have been largely performed for individuals of European ancestry. Unfortunately, results obtained using individuals of a specific ancestry usually transfer poorly to a different ancestry and are therefore not generalizable. This is due to differences in effect sizes across ancestries because of different patterns of linkage disequilibrium (LD) and allele frequencies and/or gene-by-context interactions. Third, prediction accuracy can be different even between groups of individuals of the same ancestry, stratified according to age, sex, or socio-economic status, pointing to the presence of gene-by-context interactions. My research program will focus on investigating the role of gene-by-context interactions on the genetic architecture and polygenic prediction of complex traits. My first goal is to elucidate the importance of gene-by-context interactions and their contribution to effect size differences in multi-ancestry samples. To do so, I will develop a new analytical strategy using a combination of methods that will increase the power for discoveries. This will result in a comprehensive assessment of the contribution of gene-by-context interactions to complex trait variation. My second goal is to evaluate the utility of accounting for gene-by-context interactions to improve phenotypic prediction in single-ancestry as well as multi-ancestry samples. To do so, I will use statistical methods developed for agricultural breeding that have been successful at increasing complex trait prediction accuracy in that field. Considering the specific features of multi-ancestry human data, I will also develop a prediction method that models gene-by-context interactions explicitly, while simultaneously accounting for other sources of effect heterogeneity. The long-term goal of my laboratory is to increase prediction accuracy for medically relevant traits in the general population. To do so, we need methods and analysis strategies that work well on datasets with in...