Corncob: Statistical Modeling of Microbial Abundances and Dysbiosis with Beta-Binomial Regression

• Activity Number: 432 - Novel Statistical Methods for Microbiome Data Analysis
• Type: Topic-Contributed
• Date/Time: Thursday, August 12, 2021 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Bryan D Martin* and Daniela Witten and Amy D Willis
• Companies: University of Washington and University of Washington and University of Washington
• Keywords: microbiome; statistics; relative abundance; overdispersion; high throughput sequencing; dysbiosis

Abstract

Dysbiosis, or microbial imbalance, is commonly observed in microbiomes. Many methods to study dysbiosis focus only on the compositional structure of the microbiome, despite strong evidence suggesting that both the compositional structure and the stability of the microbiome can be dysbiotic. To address this, we developed corncob, a single-taxon regression model and hypothesis testing framework. Our model uses abundance tables and sample data to make inference about the underlying microbial population from which the abundances were sampled. It is able to separately test statistical hypotheses relating to microbial imbalance in the compositional structure and statistical hypotheses relating to microbial imbalance in the stability. Thus, corncob enables users to identify both differentially abundant taxa and differentially variable taxa. It does so using a hypothesis testing framework that is statistically valid even with very small sample sizes. corncob is fast, appropriate for both rare and common taxa, able to incorporate multiple covariates, and accounts for inconsistent sequencing depths.