High-order interaction analysis in genome-wide association studies using multifactor dimensionality reduction

Gene-gene interaction (GGI) plays an important role in the causation of complex diseases, and its importance has now been well recognized through the findings of many successful genome-wide association studies (GWAS). Although many statistical methods have been introduced to address GGI analysis in GWAS, these methods have mainly focused on two-way interactions, rather than on high-order interactions. In addition, rapid advancement of biotechnology has significantly increased the number of genetic variants that are detectable, which makes an exhaustive approach unfeasible. In order to overcome the computational challenge of high-order GGI analysis using statistical approach, we develop a novel and efficient strategy called Hi-Mise; a high-order interaction analysis using the Multifactor Dimensionality Reduction (MDR) method with Interaction Set Expansion, for simultaneous identification of high-order interactions. Hi-Mise consists of second-order interaction scanning step, interaction seed initialization step, and interaction set expansion step. These steps have been computationally optimized for detection of high-order interactions. Through simulation studies using real GWAS data, Hi-Mise was shown to be capable of detecting high-order interactions with high testing balance accuracies (BAs). In addition, the application of real GWAS data showed that Hi-Mise could successfully identify multiple high-order interactions simultaneously for cases with and without marginal effects.