|
Abstract:
|
High-throughput sequencing of bacterial DNA has enabled large-scale epidemiologic studies of the human microbiome. These studies have shown associations between the microbiome and risk of several human diseases. Unfortunately, analysis of microbial abundance is complicated by its compositional nature, zero-inflation and high within-subject correlation of repeat measurements. A yet unanswered question regarding microbial abundance is how to test for concordance between groups (e.g., different tissues) with respect to other grouping variables (e.g., cancer subtype). Identifying concordant species could allow insight about tissues only accessible via surgery, by instead screening other tissues. Building on Chen & Li's approach (2016), we describe a novel framework utilizing Bayesian, Zero-Inflated Mixed Beta Regression to model relative abundance and formally test concordance via posterior distributions of group means. Viability of our method is examined using both simulations and real data collected from multiple body sites in patients with and without pancreatic cancer. Utilizing this framework, specific microbes that exhibit concordant patterns between body sites are identified.
|
