Abstract:
|
False discovery rate (FDR) control in analysis of microbiome studies is challenging due to the compositional and hierarchical structure of the data. Here, we propose a new statistical framework tackling these challenges. The method consists of three major steps. First, we model the problem of identifying important microbiota into a multiple testing problem with independent test statistics. Second, we translate the phylogenetic tree into an aggregation tree. Third, we embed the multiple testing problem into the aggregation tree and control FDR. Under mild assumptions, we prove that our method will yield a consistent false discovery proportion, and has substantially improved power compared to existing methods. We apply our method to an analysis of longitudinal microbiome data from a clinical trial in leukemia patients receiving hematopoietic cell transplantation to investigate if after-transplant care modulates changes in the microbiota.
|