Epistatic interactions (EIs) of gene loci often determine complex trait phenotypes. EIs may indicate the underlying molecular mechanisms of multifactorial traits and diseases. Yet, the inference of EIs as well as EI-based gene–gene networks remains a great challenge. Neural networks have become very successful recently into classifying complex data, having revolutionized various fields; yet their complexity does not reveal how they combine the input features and this lack of explainability limits their use on genetic data. We design EpiCID, a framework based on neural networks for discovering complicated interactions between input features (SNPs in our setting).