Advancing Gene Environment Interaction Scans through Targeted Interaction Scans through Targeted

Abstract We demonstrate how Targeted Learning (TL) can be leveraged in gene-environment interaction (GxE) discovery and estimation without sacrificing power, overcoming the limitations of traditional GxE methods that have relied on overly simple, noncausal parametric models. Our pipeline for TL GxE estimation, which we refer to as tlGxE, utilizes data-adaptive Superlearning to provide more accurate and less biased GxE estimation with the ability to adjust for high dimensional confounder sets. Simulations demonstrate that tlGxE outperforms standard methods in complex scenarios and remains competitive in simpler ones in terms of estimation bias and statistical power. We apply tlGxE to data from the UK Biobank to investigate G × smoking effects on colorectal cancer risk among n = 14, 744 individuals, where five loci on chromosomes 12, 14, and 20 are identified.