A multidimensional benchmark of confounder-aware differential abundance strategies for single- and multi-cohort microbiome studies

Covariate adjustment in microbiome differential abundance analysis improves false-discovery control but reduces sensitivity, with best-performing strategies varying by preprocessing–model pairing across 71 single-cohort benchmarks.

• Why it matters: Confounding is a pervasive challenge, and prior benchmarks have not systematically evaluated end-to-end, confounder-aware differential abundance strategies across realistic single- and multi-cohort study designs.
• What they did: The study benchmarked 71 single-cohort strategies using 2,500 real-data-informed simulations plus reference-based recovery and resampling-based consistency, then tested stability under systematic perturbations; it also evaluated 63 multi-cohort extensions with cohort heterogeneity and empirical datasets.
• The result: ANCOM-BC2_CSS and edgeR_CSS were favored for high-confidence, perturbation-stable inference, while ANCOM-BC2_count and ANCOM-BC2_TMM balanced performance; in multi-cohort settings, mega-analysis showed the strongest overall balance when individual-level data were available, and results were packaged into xConDA.

AI summary of the abstract · may contain errors · not medical advice · read the paper.