Back
Science

Researchers Develop Machine Learning Tool for Cancer Risk Assessment Using Single-Cell Data

Generated on: Last updated: 2 sources
View source

New Computational Tool Uses Single-Cell Data to Predict Cancer Survival

A new computational tool called scSurvival uses machine learning to analyze data from individual cells within tumors to predict patient survival outcomes. The tool was developed with support from the National Institutes of Health (NIH) and tested on clinical data from more than 150 patients with melanoma or liver cancer.

Development and Methodology

Researchers developed scSurvival to analyze large-scale single-cell gene expression data, which provides information on thousands to millions of individual cells within a tumor. The model was trained on single-cell datasets paired with patient survival data.

The tool's methodology involves assigning each cell a weight based on its statistical relationship to survival outcomes, filtering out less relevant cells, and averaging data from the weighted cells to generate predictions.

Reported Performance and Findings

When tested on clinical data, the scSurvival model predicted survival outcomes more accurately than traditional analytical methods, according to the researchers.

The model also traces its predictions back to specific cell groups. In the tested cases:

  • It identified specific populations of immune cells and tumor cells linked to either better or worse survival outcomes.
  • For melanoma patients, the tool identified cell populations associated with responses to immunotherapy.

Research Context and Statements

Traditional methods for analyzing tumor data often average information across entire tumors or broad cell types, which researchers state can obscure critical biological nuances present at the single-cell level.

"A risk assessment tool which identifies high-risk patients and provides clues about the underlying reasons could be helpful for difficult cancers," stated Anthony Letai, M.D., Ph.D., of the NIH's National Cancer Institute.

Corresponding author Zheng Xia, Ph.D., of Oregon Health & Science University, said the tool considers the varying influence that individual cells have on disease progression.

Research Support

The research was supported in part by the National Cancer Institute through NIH grants R01CA283171, U01CA253472, U01CA281902, and U24CA264128.