IndOR: a new statistical procedure to test for SNP-SNP epistasis in genome-wide association studies

Abstract

Epistasis is often cited as the biological mechanism carrying the missing heritability in genome-wide association studies. However, there is a very few number of studies reported in the literature. The low power of existing statistical methods is a potential explanation. Statistical procedures are also mainly based on the statistical definition of epistasis that prevents from detecting SNP-SNP interactions that rely on some classes of epistatic models. In this paper, we propose a new statistic, called IndOR for independence-based odds ratio, based on the biological definition of epistasis. We assume that epistasis modifies the dependency between the two causal SNPs, and we develop a Wald procedure to test such hypothesis. Our new statistic is compared with three statistical procedures in a large power study on simulated data sets. We use extensive simulations, based on 45 scenarios, to investigate the effect of three factors: the underlying disease model, the linkage disequilibrium, and the control-to-case ratio. We demonstrate that our new test has the ability to detect a wider range of epistatic models. Furthermore, our new statistical procedure is remarkably powerful when the two loci are linked and when the control-to-case ratio is higher than 1. The application of our new statistic on the Wellcome Trust Case Control Consortium data set on Crohn's disease enhances our results on simulated data. Our new test, IndOR, catches previously reported interaction with more power. Furthermore, a new combination of variant has been detected by our new test as significantly associated with Crohn's disease.