Comparison of Internal Model Validation Methods for Multifactor Dimensionality Reduction to Find Complex Models

Determining the genes responsible for complex human traits can be challenging when the underlying genetic model takes a complicated form, such as genetic heterogeneity (in which different genetic models can result in the same trait) or epistasis (in which genes interact with other genes and the environment).  Multifactor Dimensionality Reduction (MDR) is a widely used method that effectively detects epistasis; however, the presence of genetic heterogeneity can confound the standard cross-validation procedure used for internal model validation.  Cross-validation allows for only one “best” model and is therefore inadequate when more than one model could cause the same trait.  We hypothesize that an alternative internal model validation method, the three-way split, will be better at detecting heterogeneity models.  To test this, we will simulate genetic data that exhibits heterogeneity, implement MDR with each of the two internal model validation methods, and then compare the results.  The simulated datasets will be based on a variety of heterogeneity models (covering a range of heritabilities and penetrance models) so that the relative performance of the two internal model validation methods can be evaluated across an array of situations.  These methods will be evaluated using empirical power calculations across the various datasets.  Our results will be used to characterize the situations where in each of the two internal model validation methods is most appropriate.