An Adaptive Microbiome‐Based Truncated Test

ABSTRACT The human microbiome has been demonstrated to be associated with many complex diseases. Identifying the differences in microbial taxa across two different health conditions is clinically important, as it can enhance our understanding of disease pathology from a microbiome perspective and potentially lead to preventive or therapeutic strategies. However, there are three main challenges for analyzing microbiome data, due to compositionality, sparsity, and high dimensionality of the relative abundances. Although a few two‐sample tests have been proposed for analyzing microbiome data, the statistical power cannot be guaranteed as the true alternative hypothesis is unknown. To potentially address this issue, we propose an adaptive microbiome‐based truncated test (AMTT) that produces high power for various alternative hypotheses. Simulation studies with a wide range of scenarios are conducted, the results indicate that AMTT is not only powerful in almost all the scenarios but also effectively controls type I error rates. Real data about Parkinson's intestinal microbiome is analyzed to demonstrate its practical performance.