Colorectal cancer, the world’s second leading cause of cancer deaths, could soon be detected with far less pain, risk, and expense than traditional colonoscopies.
A team of researchers at the University of Geneva, UNIGE, has developed a stool-based test that uses artificial intelligence, AI, to map gut microbiota at the subspecies level, achieving a 90% detection rate for colorectal cancer.
The breakthrough, published in the journal Cell Host & Microbe, demonstrates that stool testing combined with advanced machine learning could become a revolutionary tool for cancer diagnosis.
With detection rates approaching those of colonoscopies—currently the gold standard but often avoided due to cost and discomfort—this new approach could transform cancer screening practices worldwide.
Colorectal cancer often goes undiagnosed until its later stages, when treatment options are limited and survival rates plummet.
Colonoscopy remains the most reliable detection tool, with a 94% accuracy rate, but uptake is low. Many patients delay screening due to invasiveness, lengthy preparation, or fear of complications.
For years, scientists have suspected that gut microbiota—the trillions of bacteria inhabiting the human digestive tract—play a crucial role in cancer development.
However, the challenge has been moving from broad observations to precise clinical applications.
Different strains of the same bacterial species can have widely varying effects, with some accelerating tumor growth and others having no impact at all.
To address this gap, the UNIGE team, led by Professor Mirko Trajkovski of the Department of Cell Physiology and Metabolism and the Diabetes Centre at the Faculty of Medicine, created the first detailed catalogue of human gut bacteria at the subspecies level.
ALSO READ: Research uncovers sugar’s hidden cancer-fighting role
“Instead of focusing only on species or individual strains, we examined microbiota at the intermediate subspecies level,” Trajkovski explained.
“This resolution is detailed enough to capture functional differences relevant to diseases like cancer, while general enough to identify patterns across populations.”
PhD student Matija Tričković, the study’s first author, described the bioinformatics challenge as massive.
“We had to design a new method for analyzing vast amounts of microbiome data. The result is the first comprehensive catalogue of human gut microbiota subspecies and a precise way.
By combining their new catalogue with clinical data, the researchers trained AI algorithms to detect cancer-linked microbial patterns in stool samples.
The results exceeded expectations: 90% of colorectal cancer cases were identified, a performance nearly equal to colonoscopies and superior to all existing non-invasive tests.
“Although we were confident, the outcome was striking,” Tričković said. “With more clinical data, this model could eventually match colonoscopy accuracy and serve as a routine first-line screening tool.”
Under this model, colonoscopies would be reserved for confirmation and further examination in patients flagged by the stool test, potentially reducing costs and avoiding unnecessary invasive procedures.
The researchers believe this innovation has applications well beyond one type of cancer.
By revealing the mechanisms through which subspecies variations influence health, the method could help develop non-invasive diagnostics for a wide range of conditions, from metabolic disorders to other cancers.
“This approach opens a new world of possibilities,” Trajkovski noted. “With one microbiota analysis, we may soon be able to detect multiple diseases without the need for invasive tests.”
A first clinical trial, in partnership with Geneva University Hospitals, is being prepared to determine which stages of cancer and types of lesions the test can identify.
Researchers are optimistic that the method will be refined to capture early-stage cancers, when treatment is most effective.
Globally, the need for improved colorectal cancer screening is urgent. Cases are rising among younger adults for reasons still not fully understood.
A simple, affordable, and non-invasive test could dramatically expand access to screening, particularly in low- and middle-income countries where colonoscopy availability is limited.
If clinical trials confirm its effectiveness, the AI-powered stool test could redefine how medicine approaches cancer detection.
It represents a convergence of microbiology, bioinformatics, and artificial intelligence in service of public health.
As Trajkovski summarized: “Food is medicine, the microbiota is a mirror of our health, and with the right tools, it can also be our diagnostic ally.”
For patients reluctant to undergo colonoscopies, the promise of a reliable, comfortable, and cost-effective alternative could make the difference between catching cancer early and facing it too late.
Science Daily
