Glossary
October 26, 2026

Scientists Build Computer Copies of Patients' Breast Tumours, and an AI Program Uses Them to Suggest Better Treatments

Researchers in London and Cambridge built computer copies of 31 patients' tumours from a hard-to-treat kind of breast cancer and used an AI program to test treatments on them. The copies matched what really happened to the patients about three times out of four, and for most patients whose cancer didn't fully respond, the program suggested an extra drug that might have helped. The suggestions have not yet been tried on patients.

Scientists at University College London and the University of Cambridge reported today that they have built "digital twins" of breast tumours: computer copies of a patient's cancer, built from the cancer's genes, that can be given different treatments on screen to see what might happen. Their study appears in the journal Cell Reports Medicine.

They focused on triple-negative breast cancer, which makes up roughly 15 percent of breast cancers and is harder to treat because it lacks the targets that many breast cancer drugs aim at. The team built copies of the tumours of 31 women who had taken part in an earlier clinical trial, so the real outcomes were already known. Each copy was given three treatments on the computer: chemotherapy alone, chemotherapy plus a drug called olaparib, and no treatment.

The copies predicted correctly whether the treatment would work in 74 percent of cases, compared with what actually happened to the patients. That also means about one prediction in four was wrong. For 13 patients whose cancer didn't fully respond to treatment, the program found at least one extra drug that might have helped in 12 of them. For 10 of those, it predicted the added drug could have cleared the invasive cancer from the breast and nearby lymph nodes.

The team used a kind of AI that follows written-out rules of logic, rather than one that learns patterns from huge amounts of data. The advantage, they say, is that doctors can follow its reasoning step by step instead of being handed an answer they can't check. The lead researcher, Professor Jasmin Fisher, said the team sees the model "as an aid for existing cancer clinical practice, not an alternative to it." Doctors might still turn down a suggestion, for example if a drug would be too harsh for a particular patient.

This is early work. It was tested on only 31 patients, entirely on the computer, using people who had already been treated. None of the suggested extra treatments has been given to a patient because of this model. The researchers present it as a step toward using tumour copies in the clinic, which will need larger studies and real trials first.

Sources

  1. UCL — AI-based computer model offers new weapon against hard-to-treat cancers (press release via EurekAlert)eurekalert.org · primary
  2. Open Access Government — Symbolic AI-driven digital twins predict personalised cancer treatmentsopenaccessgovernment.org
Follow the timeline
Get an email when new entries are added.
© 2026 Sugarpine