Researchers at the University of Geneva have crafted an AI-driven approach to anticipate the likelihood of cancer metastasis by mapping gene expression in tumor cells. Their tool, MangroveGS, leverages complex genetic signatures to assess whether colon cancer cells-and potentially others-will spread, offering a fresh angle for personalizing treatment and uncovering new drug targets. This study was published in Cell Reports.
Cancer has traditionally been viewed as disorderly cell growth, but this team suggests a more nuanced perspective: it’s a warped continuation of developmental processes. Key genetic and epigenetic programs, typically silent after early growth, get reactivated, shaping tumor formation along predictable biological pathways. Pinpointing which tumor cells break free to seed secondary growths remains a major unknown-one MangroveGS aims to illuminate.
Metastasis accounts for the majority of cancer-related deaths, especially in colon, breast, and lung cancers. Detecting circulating tumor cells usually means the disease has already spread, limiting treatment options. The Geneva researchers tackled this challenge by isolating and cloning individual cancer cells, testing their migration potential both in lab dishes and mouse models to accurately characterize their metastatic abilities.
Examining about thirty cloned cells from two colon cancer samples, the team discovered that a cell’s metastatic behavior wasn’t dictated by solitary gene changes but rather by distinctive patterns of gene activity shared across clusters of related cells. These gene expression patterns became the foundation for training MangroveGS, an AI system designed to synthesize hundreds of gene signatures, making its predictions robust against individual cell variability.
MangroveGS demonstrated almost 80% accuracy in predicting cancer recurrence and metastasis, surpassing existing prediction models. Intriguingly, the gene signatures identified in colon cancer also held predictive power for other cancers such as stomach, lung, and breast cancer. The AI can analyze tumor RNA from hospital samples and rapidly generate a metastasis risk score accessible through a secure system for clinicians.
This advancement could shift how oncologists manage therapy by avoiding overtreatment in low-risk patients while intensifying care for high-risk cases. Moreover, it promises to refine clinical trial enrollment by targeting patients who stand to benefit most, potentially speeding the development of new treatments. As AI increasingly integrates into oncology, tools like MangroveGS exemplify how genetics and machine learning can partner to decode cancer’s complex behavior.
