A comment on two-locus epistatic interaction models for genome-wide association studies

Detection of epistatic interactions in genome-wide association studies is a computationally hard problem. Many detection algorithms have been proposed and will continue to be. Most of those algorithms measure their predictive power by running on simulated data many times under various disease models. However, we find that there have been subtle differences in interpreting the meaning of existing disease models among the previous studies on detection of epistatic interactions. We elucidate those differences and suggest that future studies on epistatic interactions in GWAS state explicitly which versions/interpretations are employed. We also provide a way to facilitate setting parameters of disease models.