Topic: Contributions on Lasso Title: Detection of gene-environment effects in GWAS using logistic regression with latent exposure Authors:  Gregory Nuel - CNRS (France) [presenting]
Flora Alarcon - ParisDescartes University (France)
Abstract: The detection of gene-environment (GE) interactions is of utmost interest in genetic epidemiology, particularly as it leads to a better understanding of underlying disease etiology and subsequently the development of disease prevention and intervention strategies. However, to date, only few loci that interact with the environment have been discovered, demonstrating that the problem is very challenging due to various causes, including the presence of confounding factors. Indeed, it is very common that environment exposure is only observed through proxy covariates that often act as confounding factors and cause a drastic loss of power in GE interaction detection. We suggest accounting for these confounding factors by introducing a binary latent exposure in the logistic regression. This logistic regression with latent exposure (LRLE) model enables a significant power gain in the detection of GE interactions. The usefulness of the model is illustrated through a simulation framework where its properties and performances are studied and compared to two standard approaches for detecting GE interactions: the cases-controls and the cases-only tests. Through these simulations, LRLE proves itself to provide a dramatic power improvement over existing approaches and thus appear as a promising new method for detecting GxE interactions in GWAS.