Biological networks and epistasis in genome-wide association studies

Over the last few years, technological improvements have made possible the genotyping of hundreds of thousands of SNPs, enabling whole-genome association studies. The first genome-wide association studies have recently been completed to detect causal variant for complex traits. Although increasing evidence suggests that interaction between loci, such as epistasis between two loci, should be considered, most of these studies proceed by considering each SNP independently. One reason for this choice is that looking at all pairs of SNPs increases dramatically the number of tests (approximatively 50 billions of tests for a 300,000 SNPs data set) that faces with computational limitation and strong multiple testing correction.We proposed to reduce the number of tests by focusing on pairs of SNPs that belong to genes known to interact in some metabolic network. Although some interactions might be missed, these pairs of genes are good candidates for epistasis. Furthermore the use of protein interaction databases (such as the STRING database) may reduce the number of tests by a factor of 5,000.Results using this approach will be presented on simulated data sets and on public data sets.