Statistical Methods for Large Scale Genetic Analyses

Abstract

Population scale genomic analyses have informed the development of novel therapeutics, diagnostics, and understanding of disease etiology. Among the recent developments in human genetic association analyses, electronic health record (EHR) linked biobanks and population scale whole genome sequencing (WGS) have provided fertile ground for association discovery. In tandem with the emergence of these approaches, novel computational and statistical approaches are needed to address the methodological challenges of working with these data. In Chapter 2, I present study design recommendations and meta-analysis results for genetic association studies applied to clinical laboratory data in EHR linked biobanks. We conducted genome-wide association studies (GWAS) of 70 clinical lab traits from both the Michigan Genomics Initiative (MGI) and BioVU from the University of Vanderbilt health system. In addition to the discovery of novel association results, we conducted systematic study design analyses in parallel across the two biobanks to inform recommendations for association studies of lab traits. In Chapter 3, I present a novel sparse Mendelian randomization (MR) method for causal inference. MR methods are an instrumental variable approach for inferring the causal effect of an exposure on an outcome using genetic variants as an instrument. Under settings where the proportion of genetic variants that are causal is low, current approaches that assume dense genetic architectures may have poor statistical power. Here, we present a novel Bayesian MR method using a horseshoe prior which can be applied to summary statistics. The horseshoe prior is a continuous-scale shrinkage prior which facilitates variable selection. We use simulations to evaluate the performance of the method across genetic architectures. We apply the method to lab trait GWAS summary statistics. In Chapter 4, I present a novel method for estimating the rate at which somatic clones are expanding in clonal hematopoiesis. Clonal hematopoiesis refers to a state of mosaicism in blood defined by the acquisition of oncogenic driver mutations at an appreciate clone size and can be identified using WGS. Previous approaches for describing the growth of these mutations have relied on longitudinal sequencing methods. Here, we develop a Bayesian hierarchical model for estimating the parameters that describe the expansion of driver variants. In contrast to previous reports, our method only requires a single draw of blood. We validate the method using simulations and longitudinal amplicon sequencing. We apply our method to ~5,000 samples with clonal hematopoiesis from the Trans-Omics for Precision Medicine (TOPMed) sequencing initiative, enabling association studies of the molecular determinants of clonal expansion.