Power and sample size calculation for microbiome epidemiology

Accurately assessing statistical power as a function of sample size and effect size is critical for good study design, particularly with respect to complex human populations and high-dimensional molecular epidemiology. Microbiome data especially pose unique challenges, considering the many biological factors that can influence the microbiome, the multiple types of molecular measurements possible, and their technical and biological variability including compositionality, zero-inflation, and measurement error. Standard methods for calculating power may thus be inadequate for measuring associations between microbial features and biological variables of interest. We demonstrate this using simulated and synthetically spiked microbial profiles containing known relationships of varying types. Standard parametric or rank-based tests consistently mis-estimated power, suggesting that richer hierarchical models or simulation frameworks for study design will be more appropriate. We are currently testing such models using this benchmarking approach to provide a suite of methods for accurate feature-wise and omnibus test power calculations in human microbiome population studies.