Toward a Network-based Approach to Modeling Epistatic Interactions in Genome-wide Association Studies
In genome-wide association studies, genetic variants can be analyzed within statistical frameworks that take the underlying genetic model into consideration (e.g., the Cochran–Armitage test for trend or the logistic regression model). Should a researcher was to investigate the role played by the interaction between two or more genetic variants (epistasis), it would suffice to add interaction terms to a logistic regression model, where these terms correspond to the product of the genetic variants. However, such a model would not capture the subtlety of the interactions in the human genome. Thus, this paper proposes a network-based model that allows centrality measures from graph theory to characterize genetic variants and their interactions in a genome-wide context, and includes a simulation study that shows its applicability.