Geneva Scientists Build AI That Predicts Whether Cancer Will Spread — With 80% Accuracy Across Multiple Tumor Types

Scientists at the University of Geneva have developed an artificial intelligence tool that can predict whether a cancer will spread to other organs with approximately 80 percent accuracy — and have shown that the same underlying genetic signatures driving that prediction work not just for the tumor type the tool was trained on, but across multiple distinct cancers including stomach, lung, and breast cancer.

The tool, called MangroveGS — for Mangrove Gene Signatures — was developed by a team led by Ariel Ruiz i Altaba in the Department of Genetic Medicine and Development, with co-authors including Aravind Srinivasan and Arwen Conod. Their findings, published in ScienceDaily this week, describe how the system analyzes RNA sequencing data from tumor samples to generate a metastasis risk score that can be shared securely with oncologists and patients.

The central discovery underpinning MangroveGS is conceptually significant: cancer metastasis, long believed to be a largely random process driven by accumulated genetic mutations in individual tumor cells, is not actually random at all. Instead, the Geneva team found that tumors follow distinct biological programs — coherent patterns of gene expression that predict whether they will remain localized or aggressively migrate to distant sites in the body.

Rather than relying on the profile of a single cell, the tool "exploits dozens, even hundreds, of gene signatures" and measures the interactions between related cancer cells that form spatial clusters within the tumor microenvironment. This ensemble approach makes MangroveGS substantially more robust than earlier single-gene biomarker tests, which were often defeated by the heterogeneity of real tumors.

The team trained MangroveGS on colon cancer data and then tested it on an independent validation set, achieving roughly 80 percent accuracy in predicting both metastatic spread and recurrence following surgery. Crucially, when the researchers applied the colon-cancer gene signatures to datasets from patients with stomach, lung, and breast cancer, the predictive power transferred with only modest degradation — a result the team described as "remarkable" and suggestive of conserved biological mechanisms underlying metastatic competence across cancer types.

The practical implications are immediate. Current clinical guidelines for treating colorectal cancer after surgery rely on staging systems — based on tumor size, lymph node involvement, and imaging — that have changed little in 30 years. Oncologists often face a difficult decision: should a patient with a Stage II colon cancer receive adjuvant chemotherapy, with its significant side effects and costs, when their chance of recurrence is already low? MangroveGS could provide a molecular answer to that question, identifying high-risk patients who need aggressive follow-up while sparing low-risk patients from overtreatment.

"The research revealed that cancer metastasis follows structured biological patterns rather than being random," Ruiz i Altaba said. "This means we can potentially identify which tumors are likely to spread before they actually do — and intervene accordingly."

The tool also has implications for clinical trial design. By using MangroveGS to enrich trial populations with high-risk patients, pharmaceutical companies could conduct smaller, faster, and cheaper trials of anti-metastatic drugs — a category where the pharmaceutical pipeline has historically been thin due to the difficulty of detecting metastatic activity early enough to measure drug benefit.

The Geneva team is now seeking regulatory approval to deploy MangroveGS in Switzerland and is in early discussions with oncology groups in the United States and United Kingdom about validation studies. The technology has been licensed to a startup company, MangroveDx SA, which is developing a CLIA-compliant laboratory test that would deliver the RNA sequencing analysis and risk score through standard clinical laboratory channels.

Cancer is the second leading cause of death worldwide, accounting for approximately 10 million deaths annually. Metastatic cancer — disease that has spread beyond its original site — accounts for the overwhelming majority of those deaths, making earlier and more accurate prediction of metastatic risk one of the highest-priority targets in oncology.