Benchmarking Statistical Methods for Differential Composition Analysis in Population-Scale Single-Cell Data

LOCOM is the top-performing method for differential cell-composition analysis in population-scale scRNA-seq, controlling the false discovery rate while maintaining good statistical power across three datasets.

• Why it matters: Differential composition shifts can reflect disease pathogenesis and suggest therapeutic targets, but scRNA-seq cell counts are compositional, zero-inflated, overdispersed, and confounded, and existing methods lack systematic evaluation.
• What they did: The study benchmarks 12 differential composition methods on three population-scale scRNA-seq datasets, then uses data-informed permutations and simulations based on the real data to assess false discoveries and power.
• The result: Real-data comparisons show low agreement across methods, and most methods exhibit excessive false discoveries in permutation/simulation tests; LOCOM controls FDR best, with ANCOM-BC and centered-log-ratio linear regression next.

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