Technological and scientific advances in recent years have revolutionized genomics. For example, decreases in whole genome sequencing (WGS) costs have enabled larger WGS studies as well as larger imputation reference panels, which in turn provide more comprehensive genomic coverage from lower-cost genotyping methods. In addition, new technologies and large collaborative efforts such as ENCODE and GTEx have shed new light on regulatory genomics and the function of non-coding variation, and produced expansive publicly available data sets. These advances have introduced data of unprecedented size and dimension, unique statistical and computational challenges, and numerous opportunities for innovation. In this dissertation, we develop methods to leverage functional genomics data in post-GWAS analysis, to expedite routine computations with increasingly large genetic data sets, and to address limitations of current imputation reference panels for understudied populations. In Chapter 2, we propose strategies to improve imputation and increase power in GWAS of understudied populations. Genotype imputation is instrumental in GWAS, providing increased genomic coverage from low-cost genotyping arrays. Imputation quality depends crucially on reference panel size and the genetic distance between reference and target haplotypes. Current reference panels provide excellent imputation quality in many European populations, but lower quality in non-European, admixed, and isolate populations. We consider a GWAS strategy in which a subset of participants is sequenced and the rest are imputed using a reference panel that comprises the sequenced participants together with individuals from an external reference panel. Using empirical data from the HRC and TOPMed WGS Project, simulations, and asymptotic analysis, we identify powerful and cost-effective study designs for GWAS of non-European, admixed, and isolated populations.In Chapter 3, we develop efficient methods to estimate linkage disequilibrium (LD) with large data sets. Motivated by practical and logistical constraints, a variety of statistical methods and tools have been developed for analysis of GWAS summary statistics rather than individual-level data. These methods often rely on LD estimates from an external reference panel, which are ideally calculated on-the-fly rather than precomputed and stored. We develop efficient algorithms to estimate LD exploiting sparsity and haplotype structure and implement our methods in an open-source C++ tool, emeraLD. We benchmark performance using genotype data from the 1KGP, HRC, and UK Biobank, and find that emeraLD is up to two orders of magnitude faster than existing tools while using comparable or less memory.In Chapter 4, we develop methods to identify causative genes and biological mechanisms underlying associations in post-GWAS analysis by leveraging regulatory and functional genomics databases. Many gene-based association tests can be viewed as instrumental variable methods in which intermediate phenotypes, e.g. tissue-specific expression or protein alteration, are hypothesized to mediate the association between genotype and GWAS trait. However, LD and pleiotropy can confound these statistics, which complicates their mechanistic interpretation. We develop a hierarchical Bayesian model that accounts for multiple potential mechanisms underlying associations using functional genomic annotations derived from GTEx, Roadmap/ENCODE, and other sources. We apply our method to analyze twenty-five complex traits using GWAS summary statistics from UK Biobank, and provide an open-source implementation of our methods. In Chapter 5, we review our work, discuss its relevance and prospects as new resources emerge, and suggest directions for future research.
【 预 览 】
附件列表
Files
Size
Format
View
Statistical and Computational Methods for Genome-Wide Association Analysis
PDF
download
878987797
028-85220240
