Statistical Methods in Genetic Association Studies.

Abstract

For the past a few years, the genomewide association studies (GWAS) have rapidly grown in scale and complexity, and have provided new insights into complex disease genetics. For loci identified by GWAS, investigators are often interested in estimating the genetic effect size to help understand the genetic contribution of these loci to the disease risk or trait variation. However, estimates of the genetic effect size based on initial GWAS sample(s) tend to be upwardly biased as a consequence of “winner’s curse”. Overestimation of genetic effect size in initial studies may cause follow-up studies to be underpowered and so to fail. In Chapter 2 of this dissertation, I quantify the impact of the winner’s curse for one- and two-stage case-control association studies. I then propose an ascertainment-corrected maximum likelihood method to reduce the bias and so achieve more accurate estimation of genetic effect size. I show that overestimation of the genetic effect by the uncorrected estimator decreases as the power of the association study increases and that the ascertainment-corrected method reduces bias when power to detect association is low to moderate as expected for GWAS for complex disease. In Chapter 3, I extend this study to quantitative trait association studies and show that the results are similar to case-control association studies. Following GWAS, testing for association between gene expression and identified SNPs has the potential to help understand the relationship between these SNPs with the trait, and identify the gene(s) and variants most likely to influence the trait. In Chapter 4, I describe five testing procedures for using the allelic expression imbalance (AEI) to detect cis-acting regulatory SNPs (rSNP), focusing on the situation when the rSNP and a transcribed SNP (tSNP) are in incomplete linkage disequilibrium (LD) and there is no phase information. My simulations show that the type I error rates for all tests are well controlled, and the relative rankings of the tests depend on the LD between the two SNPs, AEI effect size of the rSNP, sample size, and allele frequencies of the SNPs.