Modeling interactions with known risk loci-a Bayesian model averaging approach

Abstract

Genome-wide association studies (GWAS) are now clearly established as a powerful method for detecting loci involved in the etiology of common complex diseases. Most diseases and traits studied using the GWAS approach now have several loci that have been shown to be convincingly replicated. It is generally the case that these loci have been identified using single locus association scans of genotyped or imputed SNPs and very few loci have been identified by taking interactions into account. We propose a method that assesses the evidence of association at each SNP by modeling the effect of the locus in combination with other known loci. We use a Bayesian model averaging approach that combines the evidence across several different plausible models for the way in which the loci interact. We show that the method has good power both when the association is the result of marginal effects only, and when interaction with a known locus occurs. The method is implemented as an option in the program SNPTEST.