Study: AI predicts which cancer patients respond to immunotherapy

A patient's tumor is a battlefield, and immunotherapy asks the body's own immune system to win it. When the drug works, the effect can last for years. When it does not, the patient has paid for a treatment that cannot help, and has lost time that could have gone to another therapy. A team at Pohang University of Science and Technology (POSTECH) has now built an artificial intelligence method that tries to tell these two outcomes apart before the first dose.

The method comes from a group led by professor Sanguk Kim of the Department of Life Sciences, with Minsoo Kim, an integrated MS-Ph.D. student at the Graduate School of Artificial Intelligence. It analyzes genetic information from cancer patients to predict their response to immunotherapy, and the results were published in npj Digital Medicine. The central target is the senescent cell, an aged cell that the authors describe as a troublemaker in the tumor's surroundings: it quietly suppresses immune cells. Clinicians find that environment hard to read case by case, which is why researchers have turned to AI.

The obstacle is data. A cancer study typically includes relatively few patients, while the genetic information to analyze per patient is enormous. An AI trained that way can memorize its training group and fail on new patients, like a student who knows the practice questions by heart. The POSTECH team changed the order of operations. It first identified tumors with a senescent microenvironment in which immune activity was suppressed, then progressively stratified patients to judge whether they were likely to respond. The authors say this stepwise approach kept the AI from becoming biased toward one patient group.

The results are the team's own. The method outperformed biomarkers currently used in clinical practice in predicting treatment response across four cancer types, including skin cancer, and six patient cohorts. Against other existing prediction methods, it also gave consistent results in an independent skin cancer cohort that was not part of its training. Sanguk Kim said that taking cellular senescence into account "could help more cancer patients benefit from immunotherapy in the future."

For oncologists, the immediate gain would be a prediction made in advance, from a patient's own genetic data, about whether immunotherapy is worth starting. The authors describe this as a step toward personalized cancer treatment, in which the same cancer may call for different treatments in different patients. The study gives no date for use in hospitals, and the team says it will run follow-up studies to model the tumor environment more precisely. Until those studies and independent use confirm the method, its benefit remains a prospect for patients, not a practice.