High-order gene-gene interaction analysis in Genome Wide Association Studies

Abstract

Genome-wide association studies already have found hundreds of associations between genetic variants and complex human diseases and traits. Most GWA-studies are concentrated in single variants effect size. In this reason, most variants which were found by studies can only explain a small part of human diseases and traits. Consequently, many researchers study gene-gene or gene-environment interactions and develop for these interactions. In 2001, Ritchie et al. proposed MDR method for the determination of gene-by-gene and gene-by-environment interactions. This method has a benefit of fitting and interpreting effects of gene-gene interactions. However this method can only be adopted for case- control data and cant adjust other environmental variables. To overcoming these disadvantages, in 2007, Xinag-Yang et al. proposed generalized MDR method. Despite of its benefits, this method also has a data sensitive problem. In other words, performance of GMDR method is affected by erroneous samples Erroneous samples means which diverge from the usual tendencies of other samples which expected to be similar with them. In this reason, we first show about negative effects of erroneous samples in GMDR by using a toy example and propose two methods for reducing effects caused by erroneous samples. Methods to reduce effects of erroneous samples which we propose are L-estimator GMDR and M-estimator. L-estimator and M-estimator are statistical methods deriving for robust estimation. In this study, to adjust concepts of L-estimator and M-estimator to GMDR method has advantages in consistency of choices. As a result we reveal that L-estimator GMDR and M-estimator has a benefit to robustness by simulation and real data analysis.