A research team from the Institute of Science Tokyo announced on Mar. 23 the development of a computational method called scSurv, which links individual cells to patient outcomes using widely available bulk RNA sequencing data. The new approach utilizes single-cell reference datasets alongside patient survival data to infer how specific cell populations within complex tissues contribute to disease risk and prognosis.
The significance of this work lies in its potential to uncover which cells drive disease progression or resistance, particularly in cancer, where understanding cellular behavior could guide future treatment strategies. Advances in single-cell sequencing have enabled researchers to measure gene expression at the level of individual cells, but comprehensive datasets that combine this information with clinical outcomes remain limited.
The scSurv model addresses this gap by using single-cell RNA sequencing data as a reference point for deconvoluting bulk tissue samples and estimating latent cell states—groups of cells sharing similar gene expression patterns. These estimated contributions are then linked to patient survival through an extended Cox proportional hazards model. Once trained, the model can estimate how more than 10,000 individual cells influence disease risk and prognosis and identify genes associated with disease progression.
The method was made available online on December 22, 2025, published in Volume 42, Issue 1 of Bioinformatics on January 13, 2026, and is accessible as an open-source Python package on GitHub and Zenodo. The research was led by Professor Teppei Shimamura and graduate student Chikara Mizukoshi from Science Tokyo's Department of Computational and Systems Biology along with Dr. Yasuhiro Kojima from the National Cancer Center Research Institute.
Using data from The Cancer Genome Atlas, scSurv successfully predicted patient survival across multiple cancers—including melanoma—and identified immune cell types such as macrophages known for their association with different survival outcomes. The researchers also mapped tumor regions affected by renal cell carcinoma according to clinical risk levels and demonstrated that their approach could be applied beyond cancer by testing it on infectious disease datasets.
"These findings suggest that scSurv may contribute to more advanced clinical outcome analysis and to the discovery of therapeutic targets," says Prof. Shimamura.
By linking individual cellular contributions directly to clinical results, scientists hope this tool will support more precise diagnostics and personalized treatments in various diseases.
