University of Geneva researchers have built an AI tool that reads tumors’ gene activity to forecast whether cancer is likely to spread. The approach could help spare low-risk patients from harsh treatment while focusing care on those who need it most.
Why do some cancers stay put while others spread through the body, turning a treatable disease into a deadly one? A team at the University of Geneva (UNIGE) believes artificial intelligence can help answer that question — and change how doctors manage cancer risk.
Working with colon cancer cells, researchers at the UNIGE Faculty of Medicine have identified patterns of gene activity that signal whether tumors are likely to metastasize, or spread to distant organs. They then built an AI tool, called Mangrove Gene Signatures (MangroveGS), that turns those molecular patterns into risk predictions for multiple types of cancer.
The work, published in Cell Reports, points toward more precise care: sparing low-risk patients from aggressive therapies while making sure high-risk patients are monitored and treated more intensively.
Rethinking what cancer is
Cancer is often described as a chaotic disease driven by out-of-control cells. That view is too simple, according to Ariel Ruiz i Altaba, a professor in the Department of Genetic Medicine and Development at UNIGE Faculty of Medicine, who led the study.
“The origin of cancer is often attributed to ‘anarchic cells’,” Ruiz i Altaba said in a news release. “However, cancer should rather be understood as a distorted form of development.”
During normal development, cells follow tightly controlled genetic programs to build tissues and organs. Many of those programs are later shut down. In cancer, genetic and epigenetic changes can switch some of them back on in the wrong place and at the wrong time, driving tumor growth.
That means tumors are not random, Ruiz i Altaba argued, but follow an altered logic. The challenge is therefore to find the keys to understanding its logic and form. And, in the case of metastases, to identify the characteristics of the cells that will separate from the tumour to create another one elsewhere in the body.
Why metastasis is so hard to predict
Metastasis is the leading cause of death in most major cancers, including colon, breast and lung cancer. Once cancer cells have broken away from the original tumor and entered the blood or lymphatic system, they can seed new tumors in vital organs.
By the time circulating tumor cells are detectable in blood tests, the metastatic process is already underway. At that stage, it is usually too late to prevent spread.
Researchers have long cataloged the mutations that help tumors form in the first place. But no single mutation has been found that reliably explains why some cancer cells go on the move while others stay put. Instead, the ability to metastasize seems to depend on complex combinations of genes and how they are switched on or off.
Studying that complexity in living cells is technically difficult.
“The difficulty lies in being able to determine the complete molecular identity of a cell – an analysis that destroys it – while observing its function, which requires it to remain alive,” Ruiz I Altaba added.
Building a living library of tumor cells
To get around this problem, the Geneva team created a kind of living library of colon cancer cells that they could test in the lab and in animals.
“To this end, we isolated, cloned and cultured tumour cells,” added co-first author Arwen Conod, a senior lecturer in the Department of Genetic Medicine and Development at the UNIGE Faculty of Medicine.
By cloning, the scientists produced many copies of individual tumor cells, creating distinct cell lines, or clones, that could be studied over time.
“These clones were then evaluated in vitro and in a mouse model to observe their ability to migrate through a real biological filter and generate metastases,” Conod added.
In other words, the team watched how well each clone could move and form new tumors in a realistic biological environment. At the same time, they measured the activity of several hundred genes in each clone.
From about 30 clones derived from two primary colon tumors, the researchers identified gradients of gene expression — patterns that varied in a continuous way — that were closely linked to how migratory and metastatic the cells were.
Their analysis suggested that metastatic potential is not just about the behavior of a single rogue cell. Instead, it depends on the combined properties and interactions of related cancer cells within a tumor.
Turning gene patterns into predictions
The next step was to turn those gene activity patterns into a practical tool. For that, the team turned to AI.
They fed the gene expression signatures into a model they developed, MangroveGS, which learns to associate particular combinations of gene activity with clinical outcomes such as metastasis and recurrence.
“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,” added co-first author Aravind Srinivasan, a doctoral student in the Department of Genetic Medicine and Development at the UNIGE Faculty of Medicine.
After training, the model reached nearly 80% accuracy in predicting whether colon cancer would metastasize or recur, according to the team — a performance they describe as clearly better than existing tools.
Strikingly, the signatures derived from colon cancer did not only work for colon tumors. The same patterns could also help predict metastatic potential in other cancers, including stomach, lung and breast cancer, suggesting that some of the molecular logic of metastasis is shared across tumor types.
What this could mean for patients
If validated in larger, independent studies, MangroveGS could fit into routine cancer care without major changes to how tumors are handled.
When a tumor is removed or biopsied, hospitals already analyze its cells and can sequence their RNA, the molecule that reflects which genes are active. With MangroveGS, those data could be sent in anonymized form through an encrypted portal, which would return a metastatic risk score to oncologists and patients.
“This information will prevent the overtreatment of low-risk patients, thereby limiting side effects and unnecessary costs, while intensifying the monitoring and treatment of those at high risk,” Ruiz I Altaba added.
That kind of stratification could help doctors decide, for example, which colon cancer patients truly need aggressive chemotherapy after surgery and which might safely avoid it.
The tool could also reshape cancer research.
“It also offers the possibility of optimising the selection of participants in clinical trials, reducing the number of volunteers required, increasing the statistical power of studies, and providing therapeutic benefits to the patients who need it most,” added Ruiz i Altaba.
By enriching trials with patients who are more likely to experience metastasis or recurrence, researchers could test new drugs more efficiently and with clearer results.
What comes next
For now, MangroveGS is a research tool, not a standard clinical test. The Geneva team’s findings will need to be confirmed in larger groups of patients and across more cancer types. Researchers will also need to work out how best to integrate AI-based risk scores with existing clinical factors, such as tumor stage and imaging results.
Beyond prediction, the gene signatures that MangroveGS relies on could point scientists toward new drug targets — genes or pathways that, if blocked, might prevent cancer cells from gaining the ability to spread.
As AI becomes more common in medicine, tools like MangroveGS highlight how combining deep biological experiments with advanced computation can reveal hidden patterns in disease. For patients facing a cancer diagnosis, the hope is that those patterns will translate into more personalized, less toxic and more effective care.
Source: University of Geneva