AI Virtual Cell ProteinTalks Predicts Drug Response in Breast Cancer

A Chinese team built ProteinTalks, an AI virtual cell for triple-negative breast cancer trained on 38 million protein measurements. The model predicts responses to 63 approved drugs and 59 combinations with 88% accuracy on unseen drugs.

• Model trained on 38 million protein measurements across 18 breast cancer cell lines
• 88% accuracy predicting protein changes for 81 unseen drugs
• Model can only evaluate two-drug combinations
• Dataset released as an open-source database for further research