Deconstructing the Genetic Basis of Complex Trait Variation

Genome-Wide Association Studies (GWAS) have proven remarkably successful in discovering genetic variants associated with human complex traits and diseases. However, the functional mechanisms of these variants remain poorly understood. Most traits are under natural selection, and their genetic basis is shaped by pop- ulation genetic processes. Consequently, integrating molecular and cellular understanding with those from population genetics can provide valuable insights. In this project, we propose new conceptual and statistical approaches that embody this interdisciplinary perspective. To implement and test our models, we will leverage large-scale biomedical datasets, including the UK Biobank and the All of Us biobank. Aim 1: Understanding what genes contribute most to complex traits. Most GWAS variants are non-coding, and their target genes are unknown, a crucial aspect for understanding biology and applications such as drug development. To bridge this gap, one set of approaches integrates GWAS with external functional data. Another strategy uses rare disrupting mutations within coding regions of genes. However, these methods have yielded disparate results. To address this, first, we will develop population genetics models to understand the types of genes prioritized by GWAS versus rare coding variants. Second, focusing on blood cell traits, we will examine how quantitative measures of gene importance based on these two approaches compare with and complement each other. Aim 2: Characterizing gene-by-gene and gene-by-environment interactions. The significance of genetic interactions in human complex traits has been debated, partly due to technical limitations as well as a lack of conceptual un- derstanding of how interactions arise. However, characterizing these interactions is crucial for various reasons, including for personalized medicine. To address challenges associated with previous methods, we present two new approaches. First, we propose a novel model-driven approach to leverage interactions for identifying key trait-relevant genes, considering how genetic and environmental perturbations are funneled through gene reg- ulatory networks. Second, we will explore interactions at the gene level, as opposed to the commonly studied variant level interactions. Aim 3: Developing approaches to improve accuracy and interpretability of polygenic scores (PGS). Interest in using PGS, genetic predictors of disease, is growing in clinical settings. However, a major limitation of PGS is their relatively modest predictive power. We aim to enhance PGS prediction by in- corporating prior functional information, with a particular focus on gene-level annotations often overlooked in current approaches. Another limitation of PGS is that, by design, they collapse all genetic data into one score, offering no insight into the underlying biological components of the disease in any given case. To address this, we propose decomposing PGS by tissue to uncover intermediate mediating processes. In summary, the successful completion of this project will enhance our conceptual understanding of complex trait genetics and will provide valuable insights for utilizing genetic findings in disease prediction and drug development.