Hypergraph supervised search for inferring multiple epistatic interactions with different orders

Abstract

Nonlinear interactive effects of Single Nucleotide Polymorphisms (SNPs), namely, epistatic interactions, have been receiving increasing attention in understanding the mechanism underlying susceptibility to complex diseases. Though many works have been done for their detection, most only focus on the detection of pairwise epistatic interactions. In this study, a Hypergraph Supervised Search (HgSS) is developed based on the co-information measure for inferring multiple epistatic interactions with different orders at a substantially reduced time cost. The co-information measure is employed to exhaustively quantify the interaction effects of low order SNP combinations, as well as the main effects of SNPs. Then, highly suspected SNP combinations and SNPs are used to construct a hypergraph. By deeply analyzing the hypergraph, some clues for better understanding the genetic architecture of complex diseases could be revealed. Experiments are performed on both simulation and real data sets. Results show that HgSS is promising in inferring multiple epistatic interactions with different orders.