Interpreting Cis-regulatory Mechanisms From Genomic Deep Neural Networks Using Surrogate Models

SQUID is a genomic DNN interpretability framework that utilizes domain-specific surrogate modelling to improve the understanding of underlying biological mechanisms in deep neural networks. It removes confounding effects and identifies consistent motifs, resulting in improved single-nucleotide variant-effect predictions. SQUID also supports quantifying epistatic interactions and providing global explanations of cis-regulatory mechanisms.