Bayesian Graphical Modeling of Microbial Community Composition

• Activity Number: 49 - Recent Advances in Statistical Inference on Graphs
• Type: Topic-Contributed
• Date/Time: Sunday, August 8, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Kurtis Shuler* and Juhee Lee and Irene Chen and Samuel Verbanic
• Companies: UCSC and University of California, Santa Cruz and UCLA and UCLA
• Keywords: Microbiome; Operational Taxonomic Unit; High-throughput Sequencing; Regression; Bayesian; Conditional Independence

Abstract

Rapid advances in sequencing technology have led to increased interest in microbiome studies characterizing microbial communities and their environments. For many microbiome analyses a key research task is to understand the microbiome as a whole, whose structure and function can be heavily affected by microbe-microbe interactions and interactions with its environment. In this work we present a Bayesian regression model with a graph (BRM-G) for count data to provide a holistic understanding of complex microbial communities. We employ a directed acyclic graph (DAG) to represent microbe-microbe interactions. Inference is summarized through moralization of the DAG. A regression component is included to provide insights into how environmental factors and experimental conditions are related to taxa abundance. A simulation study shows that, compared to BRM-G, alternative methods based on simple marginal correlations or not incorporating the interactions between microbes perform poorly in uncovering the complex interplay among microbial taxa. We also apply the model to a microbiome dataset to identify groups of related taxa in chronic wounds and healthy skin in human subjects.