Title: A robust and powerful statistical framework for differential abundance analysis of microbiome data
Authors: Jun Chen - Mayo Clinic (United States) [presenting]
Abstract: One central theme of microbiome data analysis is to identify differentially abundant taxa. The identified taxa could provide insights into disease etiology and, once validated, could be used as biomarkers for disease prevention and diagnosis. Due to the special characteristics of microbiome sequencing data, differential abundance analysis raises many statistical challenges, including modeling excessive zeros and overdispersion and taking into account the phylogenetic relationship among taxa. A robust and powerful framework is presented for differential abundance analysis of microbiome data. The framework consists of three parts (1) a fully generalized regression model based on the zero-inflated negative binomial model, which accounts for zero-inflation and overdispersion naturally and allows covariate-dependent dispersion to account for sample heterogeneity; (2) a new normalization method for zero-inflated sequencing data to address variable library sizes; and (3) a false discovery rate control procedure that integrates the phylogenetic tree to further improve the statistical power after differential abundance testing. The framework will be illustrated by using simulations as well as real data sets.