In a National Institutes of Health (NIH)-funded study, researchers developed a cancer assessment tool that can identify high-risk patients and the tumor cells linked to that risk. The model, called scSurvival, uses a machine learning framework designed to analyze large-scale data at single-cell resolution.
With NIH support, Oregon Health & Science University (OHSU) tested the model on clinical data from more than 150 cancer patients. The tool predicted survival outcomes and linked specific cell populations to higher risk.
"A risk assessment tool that not only tells you who may be at higher risk, but also provides clues as to why, could really help in these difficult cancers" said Anthony Letai, M.D., Ph.D., director of NIH's National Cancer Institute (NCI).
Every tumor contains a unique mosaic of cells presenting biological patterns that could indicate how a tumor will progress or respond to treatment. While researchers have managed to collect single-cell gene expression data from thousands to millions of tumor cells, analyzing it has been another story entirely.
Until now, researchers have applied methods that put the big picture in a blender, averaging cell data across entire tumors or cell type, both of which erase potentially critical nuances. The authors of the new study sought to devise an approach that better utilizes the rich datasets that are available, preserving their finer details.
"By taking a fine-tooth comb to single-cell data, scSurvival is able to consider the varying influence that individual cells have on disease progression and survival outcomes," said corresponding author Zheng Xia, Ph.D., an associate professor of biomedical engineering at OHSU.
To accomplish this, scSurvival assigns each cell a weight based on the degree that the cell is related to survival, filtering out information from less important cells. The model then averages the data from weighted cells together, forming its basis for survival predictions.
Researchers trained their model on single-cell datasets paired with survival data from hundreds of patients. They then tested it on clinical data from patients with melanoma or liver cancer and found it predicted outcomes more accurately than traditional methods.
The team also traced the model's predictions back to specific cell groups, identifying immune and tumor cells linked to better or worse survival. In melanoma, they identified cell populations associated with responses to immunotherapy.
The findings show that differences in cell populations shape how tumors behave and respond to treatment. Tools like scSurvival may help identify these patterns.
This research was supported in part by NCI through grants R01CA283171, U01CA253472, U01CA281902, and U24CA264128.
