Finding the Sources of Missing Heritability Using the Additive Epistatic Interaction Model: A Simulation Study

Abstract Objective Thousands of Genome-Wide Association Studies (GWAS) have been carried out to pinpoint genetic variants associated with complex disease. However, the proportion of phenotypic variance which can be explained by the identified genetic variants is relatively low. Thus, it is desirable to propose new computational models to explain the “missing heritability” problem. Results Here, we propose the additive epistatic interaction model, consisting of widespread pure epistatic interactions whose effects are additive and can be summarized by a genetic risk score. Based on a simulated genotype dataset, the additive epistatic interaction model well depicted genetic risks and hereditary patterns of complex diseases. When applied to real genotypic datasets, the additive epistatic interaction model showed potential for accurately classifying human populations from the 1000 Genomes Project, and individuals with and without diabetes from the UK Biobank database. Moreover, the model’s genetic risk score can be replaced by a deep learning model which is more resistant to noises. We suggest that the additive epistatic interaction model might help to explain the “missing heritability” problem. Source code is publicly available at https://github.com/wyp1125/additive_epistasis.