A statistical framework to identify gene-gene interactions underlying multiple dichotomous phenotypes from genotype data

Identifying gene-gene (G × G) interactions across multiple dichotomous phenotypes is challenging due to the extreme sparsity of SNP-derived interaction matrices and reduced statistical power by binary outcomes. Existing G×G association methods are restricted to either single or multiple continuous phenotypes. Here we introduce GiMat (Gene Interaction and Multiple-phenotype Association Test), a statistical framework that extends multivariate kernel regression to model G × G interactions jointly across dichotomous phenotypes, while explicitly capturing homogeneous and heterogeneous interaction effects. Extensive simulations demonstrate that GiMat controls type I error conservatively and adapts power flexibly to different types of relationships between interaction effects and phenotypes. Applied to type 2 diabetes and hypertension comorbidity in the UK Biobank, GiMat identified four previously unreported G×G interaction pairs associated with both phenotypes. This scalable framework enables robust discovery of complex genetic interactions underlying multiple correlated phenotypes.