# Scientists Create AI Tool to Predict Cancer Spread with Nearly 80% Accuracy
Researchers at the University of Geneva (UNIGE) have developed a groundbreaking artificial intelligence algorithm that can predict whether cancer will spread to other parts of the body, a process called metastasis, with remarkable accuracy. The tool, named MangroveGS (Mangrove Gene Signatures), represents a major advancement in cancer care by identifying which patients face the highest risk of metastasis and recurrence.
Understanding Cancer's Hidden Logic
Contrary to popular belief that cancer results from anarchic cells, Professor Ariel Ruiz i Altaba, who led the study, explains that cancer actually follows an orderly but distorted developmental program. Genetic and epigenetic changes reactivate biological programs that were suppressed during normal tissue development, creating tumors. The critical challenge lies in understanding why some tumor cells separate and migrate to create new tumors elsewhere in the body while others remain localized.
Metastasis remains the primary cause of death in most cancers, including colon, breast, and lung cancer. Currently, doctors can only detect metastasis after tumor cells already appear in the bloodstream or lymphatic system, by which point preventing their spread is extremely difficult. While scientists understand the mutations causing original tumors, no single genetic alteration explains why some cells migrate and others do not.
Innovative Research Methods
The UNIGE team faced a significant technical obstacle: determining a cell's complete molecular identity requires destroying it, but observing its function requires keeping it alive. To overcome this challenge, researchers isolated, cloned, and cultured tumor cells from approximately thirty clones derived from two primary colon tumors. These clones were evaluated both in laboratory conditions (in vitro) and in mouse models to observe their ability to migrate through biological filters and generate metastases.
By analyzing the expression of several hundred genes across these clones, the researchers identified gene expression gradients strongly linked to migratory potential. A crucial discovery was that accurate metastatic potential assessment does not depend on individual cell profiles but rather on the collective interactions between groups of related cancer cells.
The MangroveGS Algorithm
The team integrated their gene expression signatures into MangroveGS, an artificial intelligence model with a unique advantage: it exploits dozens or even hundreds of gene signatures simultaneously, making it particularly resistant to individual variations. After training, the algorithm achieved nearly 80% accuracy in predicting metastasis occurrence and colon cancer recurrence, significantly outperforming existing tools.
Remarkably, the gene signatures derived from colon cancer also predict metastatic potential in other cancer types, including stomach, lung, and breast cancer, demonstrating the tool's broad applicability across multiple forms of the disease.
Clinical Applications and Benefits
MangroveGS requires only tumor samples for analysis. Hospital staff can analyze cells and sequence their RNA, then quickly receive metastatic risk scores through an encrypted Mangrove portal that processes anonymized patient data. This rapid turnaround enables oncologists to make informed treatment decisions based on each patient's specific risk profile.
The clinical implications are substantial. Low-risk patients can avoid overtreatment, reducing unnecessary side effects and healthcare costs. Meanwhile, high-risk patients receive intensified monitoring and treatment when they need it most. The tool also optimizes clinical trial design by improving participant selection, reducing required volunteer numbers, increasing statistical power, and ensuring therapeutic benefits reach patients who need them most.
Future Implications
This research, published in Cell Reports and supported by the Swiss National Science Foundation and Swiss Cancer Research Foundation, opens pathways for more precise cancer care and identification of new therapeutic targets. By transforming our understanding of metastatic mechanisms from reactive to predictive, MangroveGS represents a significant step toward personalized oncology where treatment intensity matches individual patient risk. The ability to predict which tumors will spread before metastasis becomes detectable could fundamentally change how oncologists approach cancer treatment, potentially saving countless lives through earlier intervention and more targeted therapeutic strategies.