Phylogenetic Tree-based Microbiome Association Test

Abstract

Background: Microbial metagenomics data has large inter-subject variation and operational taxonomic units (OTU) for each species are usually very sparse. Because of these problems, non-parametric approaches such as Mann-Whitney U test and Wilcoxon rank-sum test have been utilized. However these approaches suffer from low statistical powers for association analyses and thus investigation on efficient statistical analyses is necessary. Objective: Main goal in my thesis is to propose phylogenetic Tree-based Microbiome Association Test (TMAT) for association analyses between microbiome abundances of each OTU and disease phenotype. Method: Phylogenetic tree reveals similarity between different OTUs, and thus was used to provide TMAT. TMAT calculates score test statistics for each node and test statistics for all nodes are combined into a single statistics by minimum p-value or Fishers combing p-value method. Results: TMAT was compared with existing methods with extensive simulations. Simulation studies show that TMAT preserves the nominal type-1 error and its statistical powers were usually much better than existing methods for considered scenarios. Furthermore it was applied to atopic diseases and found that community profiles of Enterococcus is associated.