Models and Methods for Genome-Wide Association Studies.

Abstract

Genome-wide association (GWA) studies provide an extensive assessment of common genetic variants across the human genome for disease association. However, due to variation in allele frequencies and disease prevalence across populations, combining samples from different geographic or ethnic groups may lead to spurious evidence for association or diminish the true association signals. In part one of this dissertation, I propose a novel approach to correct for population stratification that makes use of the large amount of genetic information available in a GWA study. Based on allele-sharing identity-by-state (IBS) measures, I develop similarity scores that can describe genetic similarity between individuals, and match cases and controls accordingly. Association tests can then be performed conditional on the matched case-control groups. I apply our approach to the Pritzker bipolar GWA study.

In part two, I extend our matching approach to families of arbitrary structure. I first apply similarity score-based matching to selected members from each family and then assign other family members to the same matched group. I modify a corrected chi-square test [Bourgain et al., 2003] following the Mantel-Haenszel procedure to account for correlations both between the family samples and between the matched cases and controls.

The rapid advance in next-generation sequencing technologies allows a near-complete survey of genomic regions of interest and even whole genomes, enabling more extensive genetic association studies of rare variants. As we plan such re-sequencing studies of a complex disease, it is useful to consider the range of plausible genetic models, e.g., risk allele frequency (RAF) and genotype relative risk (GRR) of rare or less common causal variants, based on results of previous genetic linkage and association studies for the trait. In part three, I compute the power to detect linkage and/or association as a function of genetic model, and summarize the range of models likely to yield results that are consistent with existing GWA and/or linkage studies.