Exploring cell-type associations of rare disease phenotypes using non-diseased cell atlases.

Rare diseases individually affect few patients but collectively impose a substantial global health burden. Many have a genetic origin, yet the cellular contexts in which disease genes exert their effects often remain unclear. Direct molecular investigation of disease-relevant tissues is often infeasible owing to small patient populations and frequent congenital or pediatric onset, limiting access to patient-derived samples. Here we investigate whether existing healthy human single-cell atlases can help identify candidate cellular contexts associated with rare disease phenotypes. Specifically, we test the hypothesis that cells expressing more genes linked to a phenotype than expected from their overall transcriptional activity may represent contexts particularly susceptible to disruption. Applied to more than 1300 phenotypes across multiple tissues, the analysis shows partial concordance with literature-derived phenotype-cell type relationships, with predictive performance reaching AUC ≈ 0.71 depending on the dataset. These findings suggest that transcriptional patterns captured in healthy single-cell atlases may contain informative signals about disease-relevant cellular contexts when the relevant cell populations are represented. At the same time, the results highlight the limitations of current reference resources and the need for continued efforts to improve single-cell atlases, phenotype-tissue mappings, and benchmarking datasets linking rare disease phenotypes to cellular contexts.