Metastasis drives the majority of cancer deaths, yet the molecular logic governing why some tumor cells migrate while others stay put has remained poorly understood. Researchers at the Université de Genève (UNIGE) have developed an AI-based predictive tool that brings that question measurably closer to an answer.
The work, published in Cell Reports, centers on a model called MangroveGS — short for Mangrove Gene Signatures — which predicts whether a cancer is likely to metastasize with approximately 80% accuracy, according to the announcement. That figure outperforms existing methods, and the model’s utility extends beyond the colon cancer cases used to build it, applying across multiple cancer types.
The research began with a conceptual reframe. Rather than treating cancer as cellular anarchy, the UNIGE team approached it as a distorted form of normal biological development. “Cancer should rather be understood as a distorted form of development,” said Ariel Ruiz i Altaba, professor in the Department of Genetic Medicine and Development at UNIGE’s Faculty of Medicine and the study’s lead researcher. Genetic and epigenetic changes, the team argues, reactivate developmental programs that are ordinarily switched off after early life — and those reactivated programs follow structured biological rules, not random ones.
To identify which rules govern metastatic behavior specifically, the researchers isolated, cloned, and cultured cells from two primary colon tumors, generating around thirty distinct cell clones. Each clone was evaluated in laboratory conditions and in a mouse model to measure its capacity to migrate through biological tissue and generate secondary tumors. The core methodological challenge, as Ruiz i Altaba noted, is that fully characterizing a cell’s molecular identity requires destroying it, while observing its function requires keeping it alive — a tension the cloning approach was designed to resolve.
Analyzing gene activity across those clones revealed clear expression patterns that corresponded tightly to each clone’s metastatic behavior. Critically, the study found that metastatic potential was not encoded in any single cell’s profile but emerged from how groups of related cancer cells interact with one another.
What MangroveGS Does Differently
Most predictive tools in oncology rely on a limited set of genetic markers, which makes them vulnerable to individual biological variation. MangroveGS was designed to sidestep that limitation by incorporating dozens to hundreds of gene signatures simultaneously. “This makes it particularly resistant to individual variations,” said Aravind Srinivasan, one of the researchers involved in the tool’s development. The gene signatures derived from colon cancer also proved predictive for metastatic risk in other cancer types, suggesting the underlying biological program is not organ-specific.
Clinical and Therapeutic Implications
The practical stakes are direct: a reliable early prediction of metastatic risk would allow oncologists to calibrate treatment intensity more precisely, sparing low-risk patients from aggressive intervention while ensuring high-risk patients receive it sooner. Beyond prognosis, the gene signatures identified in the study may point toward new therapeutic targets — molecular mechanisms that could, in principle, be disrupted to prevent spread before it begins.
By the time circulating tumor cells are detectable in blood or the lymphatic system, metastasis has typically already begun. A tool that flags risk at the genomic level, before that threshold, addresses one of oncology’s most persistent timing problems.
Photo by Jana Ohajdova on Unsplash
This article is a curated summary based on third-party sources. Source: Read the original article
