A0857
Title: Logistic-normal multinomial mediation analysis of microbiome community profiles
Authors: Kris Sankaran - University of Wisconsin (United States) [presenting]
Abstract: To design microbiome-based therapies, it is necessary to understand the consequences of interventions on the microbiome. Though randomized controlled trials have demonstrated that various interventions can impact microbiome composition, it is challenging to distinguish between potential mechanisms without further study of mediating factors. Building from work on mediation analysis in the compositional setting, a framework for Logistic-Normal Multinomial mediation analysis is introduced where the response of interest is a microbiome profile. Instantiations of this framework that posit specific zero-inflation, latent factor, and longitudinal structure are described. A zero-inflated quantiles-based simulation procedure is developed to guide model selection and calibrate inferences. It is illustrated how our workflow can discriminate between candidate causal mechanisms in a study of the effects of a mindfulness-based intervention on the microbiome. Our R package, LNMmediation, can be found online.