AI Algorithm Predicts Cancer Metastasis and Recurrence Risk

Some tumors spread to distant organs while others remain localized, making it one of the most critical unanswered questions in cancer care. Metastasis is the leading cause of death in most cancers, yet clinicians currently lack reliable ways to identify high-risk tumors before spread occurs. Genetic mutations driving tumor formation are well understood, but no single mutation explains why certain cells migrate while others do not. Researchers have now identified molecular patterns linked to metastatic behavior and developed a way to convert these signals into reliable risk predictions.

In research led by the University of Geneva Faculty of Medicine (UNIGE, Geneva, Switzerland), scientists studied tumor cells from colon cancer patients to understand what drives metastatic potential. Instead of analyzing single cells in isolation, the team isolated, cloned, and cultured tumor cells to observe their behavior while preserving their molecular identity. By measuring the expression of hundreds of genes across related cell populations, they identified gene expression gradients associated with a cell group’s ability to migrate and form metastases.

Building on these signatures, the researchers developed an artificial intelligence (AI) model called 'Mangrove Gene Signatures (MangroveGS) that integrates dozens to hundreds of gene expression patterns simultaneously. This multi-signature approach reduces sensitivity to individual variation and captures how groups of cancer cells behave collectively. The AI tool transforms RNA sequencing data from tumor samples into a metastatic risk score that can be rapidly interpreted in a clinical setting.

The model was trained and validated using data from colon cancer samples and tested for its ability to predict metastasis and cancer recurrence. The findings, published in Cell Reports, show that MangroveGS achieved close to 80 per cent accuracy, outperforming existing predictive tools. Importantly, gene signatures derived from colon cancer were also effective in predicting metastatic risk in other cancers, including breast, lung, and stomach cancer.

The approach allows tumor samples to be analyzed at the hospital level using standard RNA sequencing, with anonymized data processed through a secure digital platform. This could help clinicians avoid overtreatment in low-risk patients while enabling closer monitoring and intensified therapy for those at high risk. The researchers plan to further refine MangroveGS and expand its clinical use, including optimizing patient selection for clinical trials and identifying new therapeutic targets linked to metastasis.

"The great novelty of our tool, called 'Mangrove Gene Signatures (MangroveGS)', is that it exploits dozens, even hundreds, of gene signatures. This makes it particularly resistant to individual variations," said Aravind Srinivasan, PhD student and co-first author of the study.

Related Links:
UNIGE