Statistical and Computational Methods for Analyzing and Visualizing Large-Scale Genomic Datasets

Advances in large-scale genomic data production have led to a need for better methods to process, interpret, and organize this data. Starting with raw sequencing data, generating results requires many complex data processing steps, from quality control, alignment, and variant calling to genome wide association studies (GWAS) and characterization of expression quantitative trait loci (eQTL). In this dissertation, I present methods to address issues faced when working with large-scale genomic datasets. In Chapter 2, I present an analysis of 4,787 whole genomes sequenced for the study of age-related macular degeneration (AMD) as a follow-up fine-mapping study to previous work from the International AMD Genomics Consortium (IAMDGC). Through whole genome sequencing, we comprehensively characterized genetic variants associated with AMD in known loci to provide additional insights on the variants potentially responsible for the disease by leveraging 60,706 additional controls. Our study improved the understanding of loci associated with AMD and demonstrated the advantages and disadvantages of different approaches for fine-mapping studies with sequence-based genotypes. In Chapter 3, I describe a novel method and a software tool to perform Hardy-Weinberg equilibrium (HWE) tests for structured populations. In sequence-based genetic studies, HWE test statistics are important quality metrics to distinguish true genetic variants from artifactual ones, but it becomes much less informative when it is applied to a heterogeneous and/or structured population. As next generation sequencing studies contain samples from increasingly diverse ancestries, we developed a new HWE test which addresses both the statistical and computational challenges of modern large-scale sequencing data and implemented the method in a publicly available software tool. Moreover, we extensively evaluated our proposed method with alternative methods to test HWE in both simulated and real datasets. Our method has been successfully applied to the latest variant calling QC pipeline in the TOPMed project. In Chapter 4, I describe PheGET, a web application to interactively visualize Expression Quantitative Trait Loci (eQTLs) across tissues, genes, and regions to aid functional interpretations of regulatory variants. Tissue-specific expression has become increasingly important for understanding the links between genetic variation and disease. To address this need, the Genotype-Tissue Expression (GTEx) project collected and analyzed a treasure trove of expression data. However, effectively navigating this wealth of data to find signals relevant to researchers has become a major challenge. I demonstrate the functionalities of PheGET using the newest GTEx data on our eQTL browser website at https://eqtl.pheweb.org/, allowing the user to 1) view all cis-eQTLs for a single variant; 2) view and compare single-tissue, single-gene associations within any genomic region; 3) find the best eQTL signal in any given genomic region or gene; and 4) customize the plotted data in real time. PheGET is designed to handle and display the kind of complex multidimensional data often seen in our post-GWAS era, such as multi-tissue expression data, in an intuitive and convenient interface, giving researchers an additional tool to better understand the links between genetics and disease.PhDBiostatisticsUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/162918/1/amkwong_1.pd