Why some tumours metastasise while others remain localised is one of the most pressing questions in oncology. Researchers at the University of Geneva have now taken an important step towards answering it. By linking gene expression patterns to metastatic behaviour, they developed an AI-based tool that predicts the risk of cancer spread with high accuracy.
Using cells from colon tumours, the research team identified gene expression gradients that correlate strongly with a cell’s ability to migrate and form metastases. Rather than focusing on individual mutations, the researchers showed that metastatic potential arises from coordinated gene activity across groups of related cancer cells. Their findings have been published in Cell Reports.
Not a random process
According to study leader Ariel Ruiz i Altaba, cancer should not be seen as a random process. “Cancer should rather be understood as a distorted form of development,” he explains. Under genetic and epigenetic changes, dormant developmental programmes are reactivated, driving tumour growth and, in some cases, metastasis.
To study this process, researchers isolated and cloned tumour cells, then assessed their migratory behaviour in laboratory and animal models. Analysis of hundreds of genes across multiple clones revealed patterns that reliably distinguished highly metastatic cells from those with limited spread.
AI model
These gene signatures were integrated into an AI model called Mangrove Gene Signatures (MangroveGS). By combining dozens to hundreds of signatures, the model is robust to individual variation. After training, it achieved close to 80% accuracy in predicting metastasis and recurrence in colon cancer, thus outperforming existing tools. Notably, the same signatures also proved predictive for other cancers, including breast, lung and stomach cancer.
In clinical practice, the approach could support more personalised care. Tumour samples can be analysed locally, after which anonymised RNA data are processed through a secure portal to generate a metastasis risk score. “This information will prevent the overtreatment of low-risk patients, while enabling closer monitoring and treatment of those at high risk,” says Ruiz i Altaba. The tool could also improve patient selection for clinical trials, increasing efficiency and therapeutic impact.
RNACOREX
Last year, researchers developed RNACOREX, an open-source software tool that helps unravel the genetic regulatory networks driving cancer. Created by scientists at the Institute of Data Science and Artificial Intelligence (DATAI), part of the Cancer Center Clínica Universidad de Navarra, the tool identifies key interactions between microRNAs and messenger RNA that influence tumour behaviour. The study was validated using data from 13 cancer types in The Cancer Genome Atlas and published in PLOS Computational Biology.
RNACOREX analyses thousands of molecules simultaneously and builds interpretable regulatory networks by combining curated databases with real gene-expression data. In tests across multiple cancers, the tool predicted patient survival with accuracy comparable to advanced AI models, while offering greater transparency. By clearly showing which molecular interactions drive outcomes, RNACOREX supports survival analysis, biomarker discovery and more explainable precision oncology.
