Chinese AI Model Predicts Breast Cancer Drug Response in Lab Cells

AI model identifies potential treatments for triple-negative breast cancer using protein data

A Chinese research team from Westlake University in Hangzhou developed ProteinTalks, an AI model trained on more than 38 million protein measurements from 18 immortalized breast cancer cell types—16 of which are triple-negative. The model predicts drug efficacy for triple-negative breast cancer patients with 88% accuracy across 81 previously unseen drugs. It identified 800 proteins that change after treatment, including potential resistance markers. Tested on 3,000 clinical-stage molecules for three patient-derived samples, it found three additional molecules with improved efficacy. Validated in lab-grown cancer cells (melanoma, colorectal, lung, pancreatic), the model evaluates only two-drug combinations. Published in Nature on September 24, 2026, this prototype has not yet been tested in human patients.

This work could reduce trial-and-error in breast cancer treatment by identifying effective drug combinations early. However, it remains a lab-based prototype requiring clinical validation before human use. The model’s two-drug limitation and lack of human testing mean it does not yet address the need for personalized treatment in living patients. The full clinical impact depends on future trials to confirm safety and efficacy in humans.

Source: Singularity Hub