A Fast, Valid, and Powerful New Test for E×GWAS Interaction with Measurement Error in E

• Activity Number: 208
• Type: Invited
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Donna Spiegelman*+ and Huges Aschard and Molin Wang and Peter Kraft
• Companies: Harvard School of Public Health and Harvard School of Public Health and Harvard School of Public Health and Harvard School of Public Health
• Keywords: GWAS ; measurement error ; hypothesis test ; genetics

Abstract

A fast new test, the turbo test for E×Gene-wide Association Study (E×GWAS) interaction scans, is proposed, which will be useful when E, the environmental exposure, is continuous or ordinal, as is often the case. Observing that the null hypothesis for no E×G interaction can be expressed semi-parametrically as H_0:Cov(E,Y¦G=0)=Cov(E,Y|G=1), where Y is the binary disease outcome, we use ordinary least squares (OLS) to regress E on Y,G,and Y×G, where the latter term is the cross-product term between outcome and the binary or ordinal genotype. Since the Wald test statistic from this OLS regression arises from a simple non-iterative calculation, unlike the standard alternatives which rely on the iterative logistic regression model, computation time in a large E×GWAS will be sped up considerably. In preliminary simulations, the turbo test statistic has nominal Type I error and improved power over other standard test options for this setting. It should be noted that the notion of interaction is inherently model-dependent, and, for example, when H_0:Cov(E,Y¦G=0)=Cov(E,Y|G=1) is true, Cov(f(E),Y¦G=0)=Cov(f(E),Y|G=1) may not be true, where f(E) is some functional transformation of the measure