Perturbation-based propagation of bacterial composition uncertainty improves robustness and reproducibility of microbiome differential abundance analysis.

PARALLAX improves microbiome differential abundance robustness by propagating compositional uncertainty through per-replicate linear mixed modeling and scale perturbations, yielding fully stable taxa in 14.4% of subsampling cases.

• Why it matters: Differential abundance results are poorly reproducible across tools and study designs due to compositional structure, stochastic count uncertainty, and complex experimental designs that no single method addresses together.
• What they did: PARALLAX propagates compositional uncertainty through a structured scale perturbation grid using per-replicate linear mixed modeling, then summarizes inference with Ppost and Rscale across Monte Carlo replicates.
• The result: In repeated subsampling, PARALLAX found fully stable DA taxa in 14.4% versus 1.7% and 0% for comparators, and in two longitudinal mouse models it resolved robust, sex-specific, time-dependent genus shifts under repeated measures and multifactorial designs.