Epistasis Exploration as a Key Element in GWAS Data Analysis

Yutaka Yasui

Department of Public Health Science, University of Alberta, Canada  

Many genome-wide association studies (GWAS) for assessing the association of disease risk with SNPs resulted in odds ratio (OR) estimates that are very close to the null value of 1.0. This is true for GWAS of diseases that are known to have appreciable genetic components in their etiologies, and also for SNPs that have been repeatedly shown to be associated with the disease risk of interest. As studies with the epidemiologic case-control design, this is a highly unique feature of GWAS. In this talk, I will explore statistical explanations for the near-null OR estimates in GWAS and attribute this phenomenon/problem to the standard analytic approach that GWAS researchers have been using. Specifically, I will show both theoretically and empirically that certain forms of epistasis with very large ORs can be missed, and their ORs get attenuated towards 1.0, by the standard analytic approach of GWAS. Empirical illustrations will include epistasis with OR estimates greater than 100 and p-values less than 10^(-50). Implications of this explanation and recent related developments will be discussed.