Precise identification of cell states altered in disease using healthy single-cell references

Joint analysis of single-cell genomics data from diseased tissues and a healthy reference can reveal altered cell states. We investigate whether integrated collections of data from healthy individuals (cell atlases) are suitable references for disease-state identification and whether matched control samples are needed to minimize false discoveries. We demonstrate that using a reference atlas for latent space learning followed by differential analysis against matched controls leads to improved identification of disease-associated cells, especially with multiple perturbed cell types. Additionally, when an atlas is available, reducing control sample numbers does not increase false discovery rates. Jointly analyzing data from a COVID-19 cohort and a blood cell atlas, we improve detection of infection-related cell states linked to distinct clinical severities. Similarly, we studied disease states in pulmonary fibrosis using a healthy lung atlas, characterizing two distinct aberrant basal states. Our analysis provides guidelines for designing disease cohort studies and optimizing cell atlas use. In single-cell studies, combining healthy reference atlases and designed control datasets allows more precise identification of disease-associated cell states.

Automated summary

Joint analysis of diseased tissues and healthy reference data can reveal altered cell states. Using a reference atlas for latent space learning followed by differential analysis against controls improves identification of disease-associated cells, especially with multiple perturbed cell types. Reducing control sample numbers does not increase false discovery rates.