Bayesian Modeling of Metagenomic Sequencing Data for Differential Abundance Analysis

• Activity Number: 423 - Bayesian Microbiomics
• Type: Invited
• Date/Time: Thursday, August 12, 2021 : 4:00 PM to 5:50 PM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Qiwei Li* and Shuang Jiang and Nicole De Nisco and Andrew Y. Koh and Guanghua Xiao and Xiaowei Zhan
• Companies: The University of Texas at Dallas and Southern Methodist University and The University of Texas at Dallas and The University of Texas Southwestern Medical Center and The University of Texas Southwestern Medical Center and The University of Texas Southwestern Medical Center
• Keywords: Bayesian; Microbiome; Differential analysis; Dirichlet process; Markov random field; Colorectal cancer

Abstract

Advances in next-generation sequencing technology have enabled the high-throughput profiling of metagenomes and accelerated the study of the microbiome. Metagenomics sequencing data are summarized into a high-dimensional count table, which suffers from sample heterogeneity, unknown mean-variance structure, and excess zeros. To account for those characteristics, we propose a Bayesian hierarchical framework to identify a set of differentially abundant taxa. The bottom-level is a zero-inflated negative binomial model that links the observed counts to their latent normalized abundances and uses the Dirichlet process as a flexible nonparametric mixing distribution to model all latent factors that account for sample heterogeneity. The top-level is a Gaussian mixture model with a feature selection scheme that uses Markov random field priors to incorporate taxonomic tree information to identify discriminatory taxa at different taxonomic ranks. A colorectal cancer case study demonstrates that a resulting diagnostic model trained by the microbial signatures identified by our model in a cohort can significantly improve the current predictive performance in another independent cohort.