Digital tumour models predicted treatment response in 74% of cases — News-Medical

Researchers at University College London (UCL), together with a team from the University of Cambridge, developed computer models of tumours to predict the response to treatment for triple-negative breast cancer. When tested on data from 31 patients, the programme correctly predicted whether treatment would be effective in 74% of cases, News-Medical reports .

Digital twins of tumours For the study, scientists created digital twins of the tumours of participants in the PARTNER clinical trial using their genetic information. The models replicated the regimens used in the trial: chemotherapy alone, chemotherapy combined with olaparib, and a no-treatment scenario.

Olaparib is used in some cases of triple-negative breast cancer. This aggressive, fast-growing type of cancer accounts for around 15% of all breast cancer cases. Cells of this cancer lack receptors for oestrogen, progesterone and the HER2 protein, meaning that hormone therapy and HER2-targeted drugs do not work.

More current news is available on the UA.News Telegram channel Telegram . The model's logical reasoning The authors used symbolic artificial intelligence based on logical reasoning rather than prediction. According to the team, this allows doctors to see and verify the logic by which the programme proposes treatment options, unlike opaque “black box” models.

For 12 of the 13 patients who did not have a complete response to treatment, the models identified at least one additional targeted treatment option that could improve the outcome. For 10 of them, the models predicted such a high response that, after this treatment, invasive cancer would probably not be detected in the breast and lymph nodes.

Study lead Professor Jasmin Fisher said the system is considered a supporting tool for clinical decision-making rather than a replacement for a doctor. The findings are described in an article in the journal Cell Reports Medicine. The PARTNER trial, whose data were used for the modelling, ran from 2016 to 2024.