Bayesian Variable Selection for Multivariate Zero-Inflated Models: Application to Microbiome Count Data

TL;DR

This paper introduces a Bayesian variable selection method tailored for multivariate zero-inflated count data, effectively capturing microbial associations and dependencies in microbiome studies, outperforming univariate models in controlling false discoveries.

Contribution

The paper presents a novel Bayesian multivariate zero-inflated model that incorporates covariance structure for microbiome count data, improving variable selection accuracy over existing univariate methods.

Findings

Maintains type I error while identifying true associations.

Outperforms univariate methods in controlling false discovery rate.

Successfully applied to microbiome data revealing species associated with HIV.

Abstract

Microorganisms play critical roles in human health and disease. It is well known that microbes live in diverse communities in which they interact synergistically or antagonistically. Thus for estimating microbial associations with clinical covariates, multivariate statistical models are preferred. Multivariate models allow one to estimate and exploit complex interdependencies among multiple taxa, yielding more powerful tests of exposure or treatment effects than application of taxon-specific univariate analyses. In addition, the analysis of microbial count data requires special attention because data commonly exhibit zero inflation. To meet these needs, we developed a Bayesian variable selection model for multivariate count data with excess zeros that incorporates information on the covariance structure of the outcomes (counts for multiple taxa), while estimating associations with the mean levels of these outcomes. Although there has been a great deal of effort in zero-inflated models for longitudinal data, little attention has been given to high-dimensional multivariate zero-inflated data modeled via a general correlation structure. Through simulation, we compared performance of the proposed method to that of existing univariate approaches, for both the binary and count parts of the model. When outcomes were correlated the proposed variable selection method maintained type I error while boosting the ability to identify true associations in the binary component of the model. For the count part of the model, in some scenarios the the univariate method had higher power than the multivariate approach. This higher power was at a cost of a highly inflated false discovery rate not observed with the proposed multivariate method. We applied the approach to oral microbiome data from the Pediatric HIV/AIDS Cohort Oral Health Study and identified five species (of 44) associated with HIV infection.