Abstract:
|
Microbiome differential abundance analysis aims to identify microbial taxa, that are associated with certain biological or clinical conditions. Multivariate approaches, comparing microbiome composition at the level of taxa-set, are usually more powerful than univariate taxon-by-taxon analysis. Microbiome data are high dimensional, and often some or even most microbial taxa might not be associated with the outcome. Realizing the existing multivariate methods' susceptibility to the adverse effects of noise accumulation, we introduce Adaptive Microbiome Differential Analysis (AMDA), a novel tool, that examines the compositional equivalence of a taxa-set between two different conditions by utilizing the benefits of feature selection. AMDA is developed using data-driven learning approaches for robust performance across a wide range of scenarios. Extensive simulation studies and real data applications demonstrate, that AMDA is often more powerful than several competing methods, while preserving the correct type I error rate. Applying AMDA to a high taxonomic rank can provide a useful preliminary screening, and facilitate more targeted downstream microbiome fine mapping.
|