Answers are grounded in this paper and cite their sources.
Single-cell RNA sequencing has transformed our ability to identify cell types and explore cellular diversity. However, determining whether apparent cell subtypes represent true biological populations or artifacts of data analysis remains a major challenge. To address this issue, we developed scClustBench, an open-source benchmarking framework that evaluates cluster quality, stability, and reproducibility across multiple analytical approaches. Rather than relying on a single clustering method, scClustBench tests whether proposed cell subtypes can be consistently recovered. We applied scClustBench to a mouse retinal dataset containing bipolar cells, which are known to exhibit subtype diversity, and rod photoreceptors, for which subtype classification remains controversial. The framework consistently identified reproducible bipolar cell subtypes across analytical methods. In contrast, while some rod photoreceptor clusters appeared stable during resampling analyses, they could not be reproduced across embedding methods, suggesting that these partitions do not represent robust biological subtypes. Additional analyses using simulated and external datasets demonstrated that scClustBench reliably distinguishes meaningful cellular heterogeneity from analytical artifacts. These findings provide a practical framework for rigorous interpretation of single-cell transcriptomic data.
No takes yet. Share an insight, caveat, or question.
Yousef et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: