Machine-Learning Tool Uses Tumor Senescence Patterns to Forecast Immunotherapy Benefit
Researchers at POSTECH developed a machine-learning method that uses cancer patients' genetic data to predict immunotherapy response. The approach focuses on senescent cells in the tumor microenvironment, which can vary between patients and affect treatment outcomes. The tool may help identify likely responders before treatment and avoid ineffective therapy.
The POSTECH team was led by Sanguk Kim, with Minsoo Kim, and published its findings in npj Digital Medicine. The method examines patients’ genetic data to estimate whether immunotherapy will help, with particular attention to aging-related cells in the tumor surroundings that can dampen immune responses.
A central difficulty is that genetic datasets contain far more variables than there are patients, which can make models fit training groups too closely. To counter this, the researchers first flagged tumors with this immune-suppressing senescent setting, then sorted patients in stages. In experiments, the approach reportedly beat biomarkers now used in care for predicting treatment response.
If validated, this tool could help clinicians identify likely immunotherapy responders before treatment, sparing some patients from costly, ineffective therapy and allowing faster consideration of alternatives. It may also affect health systems by directing expensive treatments toward those more likely to benefit. Patients with cancers marked by senescent tumor microenvironments could be most directly affected, though the approach’s usefulness may depend on broader validation and data availability.