A team of researchers funded by the National Institutes of Health (NIH) has created a new machine learning model, scSurvival, capable of predicting cancer patient survival from single-cell tumor data. Tested on clinical datasets from over 150 patients primarily with melanoma and liver cancer, the model not only estimates survival outcomes but also identifies specific tumor cell populations linked to patient risk.
The model was developed and evaluated at Oregon Health & Science University (OHSU) under NIH support. Unlike traditional methods that average gene expression data across entire tumors or cell types, scSurvival analyzes data at the resolution of individual cells, preserving critical biological nuances that may influence tumor progression and treatment response.
How scSurvival works
ScSurvival employs a machine learning framework that assigns weights to each tumor cell based on its relationship to patient survival outcomes. By filtering out less relevant cells and emphasizing those more strongly associated with risk, the model constructs a weighted average that informs survival prediction. This approach allows for a more precise assessment of how heterogeneous cell populations within tumors contribute to disease trajectory.
Using training datasets that coupled single-cell gene expression profiles with patient survival information, the researchers validated scSurvival’s predictive power on clinical data from melanoma and liver cancer patients. The model outperformed traditional bulk-analysis methods, offering improved accuracy in forecasting survival rates.
Linking cells to survival and treatment response
Beyond prediction, scSurvival can trace its survival assessments back to particular cell groups. In melanoma cases, it identified immune and tumor cell populations associated with better or worse survival and recognized cell types linked to responses to immunotherapy. These findings underscore how diverse cell populations within tumors influence both disease development and treatment efficacy.
Why it matters
This AI model represents an advance in personalized cancer risk assessment by leveraging complex single-cell data without losing important detail. Its ability to pinpoint cell populations tied to patient outcomes could guide treatment decisions and help identify patients more likely to benefit from specific therapies, such as immunotherapy.
Background
Single-cell gene expression profiling can capture the complexity of tumor cellular makeup, but traditional analysis often averages these details into broad summaries that obscure important distinctions. By developing a tool that fully harnesses single-cell datasets, researchers aim to improve understanding of tumor biology and enhance clinical prognostication.
The study, published in Cancer Discovery, was supported by the NIH’s National Cancer Institute through multiple grants (R01CA283171, U01CA253472, U01CA281902, and U24CA264128). The National Cancer Institute leads NIH’s national effort to advance cancer research and improve patient outcomes.
Sources
This article is based on reporting and publicly available information from the following source:
Read more Health & Public Health stories on Goka World News.