Investigating statistical power of differential abundance studies

Identifying microbial taxa that differ in abundance between groups (control/treatment, healthy/diseased, etc.) is important for both basic and applied science. As in all scientific research, microbiome studies must have good statistical power to detect taxa with substantially different abundance between treatments; low power leads to poor precision and biased effect size estimates. Several studies have raised concerns about low power in microbiome studies. In this study, we investigate statistical power in differential abundance analysis. In particular, we present a novel approach for estimating the statistical power to detect effects at the level of individual taxa as a function of effect size (fold change) and mean abundance. We analyzed seven real case-control microbiome datasets and developed a novel method for simulating microbiome data. We illustrate how power varies with effect size and mean abundance; our results suggest that typical differential abundance studies are underpowered for detecting changes in individual taxon.