New AI Tool Predicts Cancer Spread With 80% Accuracy, Eliminating Need for Animal Testing

One of the most persistent mysteries in oncology is why some tumors remain localized while others aggressively spread throughout the body. Metastasis—the migration of cancer cells to new organs—is responsible for the vast majority of cancer-related deaths, particularly in colon, breast, and lung cancers. For decades, doctors have relied on broad statistical models and invasive testing to guess which tumors might migrate. Now, researchers at the University of Geneva (UNIGE) have developed an artificial intelligence system that solves this biological puzzle with unprecedented precision.[1]

The new AI tool, named MangroveGS (Mangrove Gene Signatures), analyzes the complex genetic activity of a tumor to predict its likelihood of spreading. By examining the RNA of cancer cells, the system calculates a metastatic risk score with nearly 80% accuracy, significantly outperforming existing clinical prediction methods. The breakthrough, recently published in the journal Cell Reports, represents a major leap forward in personalized medicine.[1]

To build the system, the UNIGE team, led by Professor Ariel Ruiz i Altaba, had to rethink the fundamental nature of cancer. Rather than viewing tumors as a random, anarchic growth of rogue cells, the researchers approached cancer as a "distorted form of development." They discovered that tumors reactivate specific biological programs that were originally used during the early stages of human development but were supposed to remain dormant.[2]

By cloning cells from colon tumors and measuring the activity of hundreds of genes simultaneously, the team identified specific gene expression patterns that strongly correlated with a cell's capacity to break away and migrate. They found that metastatic risk isn't determined by a single rogue cell, but rather by the collective interactions among groups of related cancer cells within the tumor.[1][2]

The researchers then fed these complex genetic signatures into a custom artificial intelligence model. According to Aravind Srinivasan, a PhD researcher and co-first author of the study, MangroveGS is unique because it processes dozens or even hundreds of gene signatures at once. This high-dimensional analysis makes the AI highly resilient to individual genetic variations, allowing it to spot the underlying "program" of metastasis even when a patient's specific mutations differ.[1]

While the system was originally trained on colon cancer data, its predictive power extends far beyond a single disease. The biological programs that drive metastasis appear to be shared across multiple cancer types. In testing, the gene signatures identified by MangroveGS successfully predicted the risk of tumor spread in stomach, lung, and breast cancers, proving that the AI has uncovered a universal mechanism of cancer migration.[1]

The clinical implications of this technology are profound. In the near future, MangroveGS could become a standard part of routine cancer care. After a biopsy, a hospital would simply sequence the tumor's RNA and upload the anonymized data to an encrypted Mangrove portal. The AI would instantly return a metastatic risk score to the patient's oncologist, fundamentally altering the treatment roadmap.[2]

For patients, this means an end to the "one-size-fits-all" approach to cancer therapy. By accurately identifying low-risk tumors that are unlikely to spread, doctors can safely scale back treatments, sparing patients from the severe side effects and financial burdens of unnecessary chemotherapy. Conversely, patients identified as high-risk can be immediately fast-tracked for aggressive, targeted therapies and intensified monitoring before the cancer has a chance to migrate.

Beyond its direct impact on patient care, MangroveGS is also transforming how cancer research is conducted. Historically, scientists have relied on implanting human tumor cells into mice to observe whether the cancer would metastasize—a slow, expensive, and ethically fraught process. Because MangroveGS can predict metastasis directly from human tumor data, it effectively eliminates the need for these animal models.[3]

In recognition of this achievement, the UNIGE team was awarded the 2026 3R Prize on June 9. The prestigious award honors scientific innovations that "reduce, refine, or replace" animal testing. The prize jury praised the algorithm as a major innovation in biological modeling, noting that it successfully bridges the gap between ethical research practices and direct clinical benefits for patients facing life-threatening diagnoses.[3]