B0775
Title: Generative mediation models for microbiome data analysis
Authors: Kris Sankaran - University of Wisconsin (United States) [presenting]
Abstract: Mediation methods add nuance to causal inferences, making it possible to attribute causal effects to specific intermediate changes. Recently, these methods have been applied to microbiome studies, where a mechanistic understanding has otherwise proven elusive. Strategies are introduced for adapting generative models of microbiome data -- including topic, zero-inflated, and logistic-normal-multinomial models -- to support mediation analysis. We will investigate the properties of these methods using a semi-synthetic simulation study and a posterior predictive visual analysis. Finally, we will study longitudinal data from a randomized control trial on the effects of a mindfulness intervention on gut microbiome composition, drawing mediators from host physiological, behavioral, and diet measurements. All code needed to reproduce the results is available in an accompanying R package.
