ProClust provides reliable statistical significance of clusters for single-cell transcriptomic data

ProClust provides p-value–based statistical significance testing that avoids false subdivisions under single-population nulls while recovering true scRNA-seq structure across varying population counts.

• Why it matters: scRNA-seq clustering can overpartition a real cell population into spurious subclusters, so biological claims need a principled way to judge whether clusters are truly separable.
• What they did: ProClust models each cluster as a log-concave distribution and uses projection-based uncertainty quantification to compute p-values, then merges clusters when the single-cluster null cannot be rejected.
• The result: In parametric and PBMC-based semi-synthetic simulations and in Human Lung Cell Atlas, Tabula Muris, and Hydra datasets, it yields clustering solutions that agree with reference annotations while retaining supported substructure.

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