On quantitative issues pertaining to the detection of epistatic genetic architectures

Converging empirical evidence portrays epistasis (i.e., gene-gene interaction) as a ubiquitous property of genetic architectures and protagonist in complex trait variability. While researchers employ sophisticated technologies to detect epistasis, the scarcity of robust instances of detection in human populations is striking. To evaluate the empirical issues pertaining to epistatic detection, we analytically characterize the statistical detection problem and elucidate two candidate explanations. The first examines whether population-level manifestations of epistasis arising in nature are small; consequently, for sample-sizes employed in research, the power delivered by detectors may be disadvantageously small. The second considers whether gene-environmental association generates bias in estimates of genotypic values diminishing the power of detection. By simulation study, we adjudicate the merits of both explanations and the power to detect epistasis under four digenic architectures. In agreement with both explanations, our findings implicate small epistatic effect-sizes and gene-environmental association as mechanisms that obscure the detection of epistasis.