Oregon Health & Science University (OHSU) researchers have developed a first of its kind tool, scSurvival, that directly links information from individual tumor cells to patient survival outcomes, allowing clinicians to understand which specific cells are driving disease progression rather than treating them all the same.
“Traditional survival models in cancer rely on bulk data, which average signals across millions of cells and obscure important heterogeneity,” explained senior author Zheng Xia, PhD, associate professor of biomedical engineering in the OHSU School of Medicine and a member of the OHSU Knight Cancer Institute. “Tumors are highly complex ecosystems where different cell subpopulations can have very different and sometimes opposing effects on patient outcomes.”
He told Inside Precision Medicine that “scSurvival is designed to directly model survival using single-cell data, preserving this heterogeneity. Instead of treating a tumor as a single entity, it treats it as a collection of individual cells and learns which specific subpopulations are most associated with survival outcomes. This enables both more accurate prediction and deeper biological insight.”
The tool was designed using a statistical method known as an attention-based multiple-instance Cox regression framework, which constructs survival prediction models from single-cell cancer cohort data while simultaneously identifying cell subpopulations that are strongly associated with patient risk.
“The attention mechanism preserves cellular heterogeneity within each patient, allowing [cell] subpopulations with higher attention scores to be more closely linked to survival probability,” the researchers explain in Cancer Discovery. “The resulting outputs of scSurvival are the attention-adjusted hazard score for each cell along with patient-level risk scores.”
Xia and team tested the performance of scSurvival in two cohorts that included 32 patients with melanoma and 124 patients with liver cancer. Together, the cohorts provided single cell RNA sequencing data for more than 1.1 million individual cells.
They found that key immune cell types were enriched for higher- or lower-hazard cells. For example, monocytes/macrophages were enriched for high-risk subpopulations in both the melanoma and liver cancer cohort, but B cells were enriched for low-risk subpopulations in the melanoma cohort and high-risk subpopulations in the liver cancer cohort.
In both groups, the tool accurately predicted patient outcomes, with cells taken from melanoma patients who did not respond to immunotherapy having significantly higher hazard scores than those taken from responders.
Xia noted that the information scSurvival provides has several translational applications. “Differential gene expression between high- and low-risk cells can be used to develop prognostic biomarkers,” he said. “Pathways enriched in high-risk populations may reveal actionable therapeutic targets, while the abundance of specific cell types can support patient stratification for treatment selection. Importantly, these insights are derived at single-cell resolution, providing greater biological precision than bulk approaches.”
At present, scSurvival is primarily a research tool but Xia believes that longer term, it has potential clinical relevance. “For example, signatures derived from survival-associated cell populations could be translated into more practical assays (e.g., bulk RNA or targeted panels) for patient stratification,” he suggested. “However, direct clinical deployment would require further validation, simplification, and standardization.”
According to Xia, one of the biggest challenges to widespread adoption of the tool is the limited availability of large, well-annotated single-cell datasets with matched survival data, as single-cell sequencing is not yet routine in clinical workflows. But as more clinical trials adopt single-cell sequencing, he expects scSurvival to see broader use in resolving disease at cellular resolution.
The investigators now plan to extend the framework to incorporate spatial transcriptomics, which will allow them to account for how cells are organized within the tumor microenvironment. “We also aim to improve the model’s robustness across datasets and sequencing platforms, and to enhance its biological interpretability. Ultimately, we hope to translate the survival-associated signatures identified by scSurvival into clinically practical tests,” Xia said.
The study findings were also presented at the American Association for Cancer Research Annual meeting 2026 and the open-source scSurvival program and its tutorials are freely available at GitHub, Zenodo and Code Ocean.
The post Single-Cell Survival Modeling Tool Offers New Precision in Cancer Prognosis appeared first on Inside Precision Medicine.
