Researchers from University College London, in collaboration with the University of Cambridge in the UK, have developed an artificial intelligence model designed to predict how tumours will respond to different treatments and suggest additional treatment options for patients who do not respond to standard therapies. The research focuses on triple-negative breast cancer, a type of cancer that is particularly difficult to treat.
اضافة اعلان
The researchers explained that the model creates virtual versions of patients’ tumours using detailed genetic data to simulate the effects of different treatments and identify options that may produce the best response. The findings were published on Monday in the journal Cell Reports Medicine.
Triple-negative breast cancer is a type of breast cancer that lacks three common receptors used to guide treatment: oestrogen receptors, progesterone receptors and human epidermal growth factor receptor 2 (HER2). The absence of these receptors limits the use of hormone therapy and HER2-targeted treatments, making the disease more difficult to treat. It is often treated with surgery and chemotherapy, while other options may include immunotherapy or additional targeted treatments, depending on the tumour’s characteristics and the patient’s condition.
Computer models are already used in cancer treatment, but advances in artificial intelligence offer new possibilities for personalised care. However, the unclear workings of some systems, often referred to as “black boxes,” have raised questions among doctors about the basis for their recommendations.
To address this issue, the researchers used symbolic artificial intelligence, which relies on logic and reasoning and allows doctors to trace and examine the steps through which the model reaches its conclusions.
To test the model’s performance, the team created virtual tumours simulating those of 31 patients who had participated in a clinical trial, using each patient’s genetic data. They then ran computer simulations comparing three scenarios: chemotherapy alone, chemotherapy combined with olaparib, a drug used in some cases of triple-negative breast cancer, and no treatment.
By comparing the simulation results with the actual outcomes of the clinical trial, the researchers found that the model correctly predicted treatment outcomes in 74 per cent of cases.
The model’s capabilities extended beyond predicting responses to the treatments tested. It also suggested additional targeted treatments for patients who did not achieve a complete response. Among the 13 patients who did not respond adequately to treatment, the model identified a potential additional treatment option for at least one patient in 12 of the cases, according to the simulations.
The results predicted that adding the proposed treatment for 10 of these patients could lead to a sufficient response to ensure that no invasive cancer remained in the breast tissue and lymph nodes after treatment.
However, the researchers stressed that these findings were based on computer simulations and did not prove that the proposed treatments would be effective in patients in clinical practice.
The team believes that this approach could eventually help support treatment decisions, particularly for patients with difficult-to-treat cancers, by identifying genes or proteins that could be targeted to make tumours more sensitive to available drugs.
The model is not intended to replace doctors, but rather to provide them with additional treatment ideas that can be evaluated. A doctor may reject a recommendation because of its potential side effects or view it as an opportunity to explore an option they had not previously considered.
Nevertheless, the widespread use of such models will require further clinical validation to confirm their accuracy and ensure that treatment decisions based on their recommendations are safe, according to the team.
