Title: Testing statistical interactions between microbiome community profiles and covariates
Authors: Michael Wu - Fred Hutchinson Cancer Research Center (United States) [presenting]
Abstract: Microbiome profiling studies are being conducted to find associations between bacterial taxa and a wide range of different outcomes. However, the dimensionality, compositionality, inherent biological structure, and limited availability of samples pose significant challenges. Community level analysis, wherein the entire profile is assessed for association with outcomes, can resolve some of these difficulties but does not easily generalize to analyzing effect modification due to bias incurred in estimating main effects. Thus, under the semi-parametric kernel machine testing framework, we propose a new framework for interaction testing at the community level that incorporates bias reduction approaches in estimating main effects while flexibly capturing interaction terms. Simulations and real data analyses show that our approach correctly controls type I error while maintaining power under a range of situations.