A variety of methods can be used to investigate how multiple genetic variants contribute jointly to a dichotomous trait. We chose to use the restricted partition method (RPM), a statistical approach designed to identify combinations of qualitative genetic and environmental factors (e.g. genotypes, categorical or dichotomized environmental exposures) contributing to a quantitative or binary trait (Culverhouse et al. 2007; Culverhouse et al. 2004) and logistic regression modeling, a more traditional statistical approach. The RPM is agnostic regarding a specific genetic model (e.g. additive, dominant, or recessive in the case of a single SNP) and was specifically designed to be sensitive even if the contribution from a combination of factors is predominantly presented as an interaction displaying little or no marginal effects. It is an exploratory method that uses the data to determine the number of distinct risk classes, with the aim of determining if modeling predictors jointly accounts for more of the variation in phenotype than summing their individual effects. It does not specifically identify interactions or test them for significance.
