What if researchers could explore how an individual tumour might respond to treatment before testing a new approach in a patient?
Scientists at UCL and the University of Cambridge are investigating that possibility through digital twins: computer models designed to represent aspects of a person’s cancer.
UCL’s report, published on 26 October, describes encouraging findings in triple-negative breast cancer. The work suggests a way to generate more focused treatment hypotheses, while leaving an important gap between a virtual prediction and a safe, effective treatment.
What Is a Digital Tumour Twin?
A digital twin is not a second tumour growing inside a computer. It is a mathematical representation built from biological information, intended to capture features that affect how a tumour behaves.
Researchers can use that model to explore questions about treatment response and possible vulnerabilities. The hope is to narrow the range of ideas that need further investigation.
That could be particularly valuable in difficult-to-treat cancers, where an option that works for one person may not work for another.
But a model is always an approximation. It cannot automatically reproduce every interaction within a tumour, its surroundings or the rest of the patient’s body.
What Did the Researchers Find?
The team created digital twins for 31 patients with triple-negative breast cancer. According to UCL, the models predicted treatment response with 74% accuracy.
For 12 of the 13 patients whose tumours had not responded completely, the models also suggested potential alternative treatment strategies.
Both findings are interesting. Neither means that AI successfully treated those patients.
The 74% figure describes prediction performance in this study; it is not a cure rate, a survival improvement or a guarantee of accuracy for a future patient. The alternative strategies were proposed computationally, not shown to improve those patients’ outcomes.
Why This Could Still Matter
Cancer research involves more possible combinations and biological mechanisms than scientists can easily test one by one.
A model that helps researchers prioritise sensible options could make that work more focused. It might also help explain why apparently similar tumours respond differently.
The useful role for AI here is not to replace the oncologist. It is to support investigation: identify a possibility, explain the reasoning where possible and give researchers something specific to test.
What Would Count as the Next Step?
Thirty-one patients is a small basis for broad claims. Larger and independent evaluations are needed to establish how well the approach generalises.
Suggested alternatives also need biological testing and, where appropriate, clinical studies. A treatment could look effective in a simulation yet fail because of toxicity, an incomplete model or other factors the computer did not capture.
This is therefore promising British research, not an announcement that digital twins are ready to choose everyone’s cancer treatment or are routinely deployed across the NHS.
There is enough here to be interested without overstating it: researchers are developing a more individual way to explore difficult cancers, and the next test is whether that leads to better decisions and outcomes in practice.
University report: UCL: AI-based computer model offers new weapon against hard-to-treat cancers ↗ — 26 October 2026. The numerical findings above follow UCL’s account.
Associated paper: Tumor digital twins uncover therapeutic strategies in triple-negative breast cancer, Cell Reports Medicine ↗. The DOI resolves to the named paper; the full publisher text was not available for independent examination for this article.
Less hype. More real-world AI progress.