Two-way interaction analysis of obesity trait from Korean population using generalized MDR

Abstract

Most common complex traits such as obesity, hypertension, diabetes, and cancers are known to be associated with multiple genes, environmental factors, and epistasis. Recently, the development of advanced genotyping technologies allows us to perform the genome-wide association studies (GWAS). For detecting the effects of multiple genes on complex traits, many approaches have been proposed for GWAS. Multifactor dimensionality reduction (MDR) proposed by Ritchie et al. (2001) is one of the powerful methods for detecting epistasis, which detects high order interactions among genes. However, MDR is computationally intensive due to its requirement of testing all possible n-way combinations. Thus, it is not practically feasible to apply MDR directly to analyzing large-scale GWAS data. We propose an efficient strategy to perform MDR analysis for GWAS data. First, select a subset of genetic factors such as SNPs with some marginal significance. Second, perform MDR analysis only for the selected genetic factors. Third, provide MDR summary results through network graph. Even for the MDR analysis for the selected SNPs, a fast computing system is required. We developed our own GPU based system for the large-scale GWAS data. We applied our strategy to Korean Association REsource (KARE) (8,838 individuals with 101,837 SNPs) for detecting two-way interactions of genetic factors associated with body mass index (BMI). We identified several genes and pathways which have been known to be associated with obesity.