Statistical Methods for Transcriptome-Wide Association Studies in Ancestrally Diverse Populations

Abstract

Transcriptome-wide association studies (TWAS) have become more commonly used in recent years. TWAS integrate genome-wide association studies (GWAS) with gene expression mapping studies in order to identify genes whose gene expression is associated with the phenotype. The main goals of TWAS are in providing insights into biological mechanisms underlying disease etiology and in helping interpret the results of GWAS. TWAS conducted in large-scale ancestrally diverse cohorts face multiple challenges, including the presence of population structure, known or cryptic relatedness and heterogeneity in phenotypic distributions across subgroups. There is a dearth of statistical methodology available to researchers that addresses the aforementioned issues. In this dissertation, we evaluate the performance of existing TWAS methods in ancestrally diverse populations and identify their limitations. We then develop new statistical methodology that addresses these limitations. We validate the performance of the novel methods in extensive series of simulations as well as in applications to large cohorts of ancestrally diverse populations from the Trans-Omics for Precision Medicine (TOPMed) program.