On Considering Epistasis in Genome Wide Association Studies

In the last decade the endeavor of gaining a better understanding of how genotypes act and interact to form the phenotypes that are disease has experienced an information explosion, brought about by the successful completion of the human genome sequencing project and advances in genotyping technologies. The high volume data produced by these new technologies gave geneticists the opportunity to study the genome from broader perspectives and helped usher in the era of the Genome-Wide association Study (GWAS). These studies allow for hypothesis-free probing of the whole genome in order to find genetic variants that may be associated with disease risk, and as such are attractive for studying complex diseases where genotype to phenotype links are not well understood.;Until recently most researchers have focused mainly on finding additive (main) effects that contribute to disease phenotype. However, it has now been shown that epistatic effects (gene-gene interactions) as well as gene-environment interactions may also play significant roles in determining phenotype. Epistasis is now of special interest since the amounts of genetic variation explained by main effects found from GWAS have so far only accounted for small amounts of variation present in the diseases or traits of interest. In order for the study and detection of epistatic effects to become as common place as main effect testing is in GWAS, more methods have to be developed and improved upon to better handle the statistical and computational challenges that dealing with millions of interactions present.;The focus of this work is on developing and improving upon methods that can detect epistatic interactions in genotypic data and improve their utility for GWAS, by a process of prioritization and filtering. I also investigate the applicability of using a covariate to adjust for confounding caused by population stratification with a view towards increasing power to detect interaction effects in diverse samples when using Multifactor Dimensionality Reduction (MDR), a popular interaction testing method. Finally I present the results from a candidate-gene study of extrapulmonary tuberculosis as well as a genome-wide association study of the same disease. The findings from the following chapters, demonstrate that in order to have efficient tests for epistasis on a GWAS scale successful methods would need to not only detect interactions but should also have the capability of filtering and prioritizing results in order to increase their adoption by researchers. I also find that population stratification is a major source of confounding for detection of interaction with MDR and that the consequences of not compensating for it are more severe for interactions than for main effects. The results from the Extrapulmonary Tuberculosis studies also reiterate the seriousness of the population stratification problem, and highlight other differences between the candidate-gene and GWAS approach.