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

• Activity Number: 526
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Holly Sweeney*+ and Jeffrey Joseph Gory
• Companies: North Carolina State University and North Carolina State University
• Keywords: multifactor dimensionality reduction ; heterogeneity ; epistasis ; cross-validation ; three-way split ; internal model validation

Abstract

Determining the genes responsible for complex human traits is challenging when the underlying genetic model takes a complicated form such as genetic heterogeneity (where different genetic models can result in the same trait) or epistasis (which involves gene-gene interactions). Multifactor Dimensionality Reduction (MDR) effectively detects epistasis, but heterogeneity confounds the standard cross-validation procedure used for internal model validation. Another internal model validation method, the three-way split, is more efficient than cross-validation and equally effective at detecting epistasis models. We tested whether or not the three-way split is better at detecting heterogeneity models by simulating genetic data (covering a range of heritabilites and penetrance models) that exhibits heterogeneity, implementing MDR with each internal model validation method, and performing empirical power calculations. We found that the three-way split's performance is just as limited as, and in some cases worse than, that of cross-validation in the presence of heterogeneity. Our results show that using a three-way split with MDR is not a good option when genetic heterogeneity is present.