Machine learning uncovers senescence signatures that predict lung adenocarcinoma outcomes

Scientists at the University of Toronto have applied machine learning to lung adenocarcinoma tissue samples and pinpointed a set of genes tied to cellular senescence that reliably forecast patient outcomes. The work, reported this week in a preprint, analyzed over 1,700 tumors from public databases and found that the expression of 15 senescence-related genes could stratify patients into high- and low-risk groups. Senescence, the state where cells stop dividing but remain metabolically active, has long been linked to cancer, but its exact role in lung adenocarcinoma has remained murky.

The team trained several algorithms on gene expression data, settling on a random forest model that achieved an area under the curve of 0.86 in predicting five-year survival. They then validated the signature in three independent patient cohorts, where it consistently outperformed clinical variables like tumor stage and age. The 15 genes include well-known players such as CDKN1A and TP53, but also less-studied members like LMNB1 and SERPINE1, which may offer new avenues for therapy.

Lung adenocarcinoma is the most common subtype of lung cancer, accounting for roughly 40% of all cases, and it kills more than 100,000 people annually in the United States alone. Current staging systems rely on anatomical spread, which often fails to capture the biological aggression of a tumor. The new signature appears to fill that gap, offering a molecular readout of how a patient's tumor is likely to behave.

The researchers also used the machine learning model to probe which senescence pathways are most active in aggressive tumors. They found that a subset of patients with high-risk signatures showed elevated activity in the p53 and retinoblastoma pathways, both of which can trigger senescence but are often disrupted in cancer. Drugs that selectively induce senescence in tumor cells, such as CDK4/6 inhibitors, are already in clinical trials for other cancers, and this work suggests they might be worth testing in lung adenocarcinoma patients with these specific signatures.

The findings are still preliminary, as the study has not yet undergone peer review. The authors caution that the signature needs prospective validation in clinical trials before it can be used to guide treatment decisions. They plan to test the model on fresh tumor samples collected in a prospective cohort starting this fall, and they are also exploring whether the signature predicts response to immunotherapy, which would expand its utility in the clinic.

For oncologists, the appeal is clear: a cheap, tissue-based test that could refine prognosis and identify patients who might benefit from senescence-inducing drugs. If the signature holds up, it would join a growing list of molecular markers, like EGFR and ALK mutations, that are reshaping how lung cancer is treated. The next step is to move from the computer screen to the clinic, where the real question will be whether these 15 genes change how patients actually fare.