AI-rewritten: This is a summary of an article from Singularity Hub, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

A Chinese research team developed an AI system called ProteinTalks to predict how patients with triple-negative breast cancer will respond to different drugs. Unlike traditional methods that rely on gene activity snapshots, this virtual cell model focuses directly on protein changes measured before and after drug treatments. The researchers analyzed over 38 million protein data points from 18 types of breast cancer cells treated with various FDA-approved drugs. This massive dataset allowed the AI to identify specific proteins that act as early signals of a drug’s effectiveness and flag potential mechanisms for tumor resistance.

The model demonstrated strong performance by predicting protein changes in unseen drugs with 88 percent accuracy. When tested on combinations of three drugs, ProteinTalks successfully identified treatments that worked better than standard therapies in lab settings using patient-derived cells. The study authors note this is the first time a virtual cell model has been tested in a clinical scenario. However, the team emphasizes that these predictions still require testing in animal models and human clinical trials before they can be used for real patients.

While ProteinTalks currently evaluates only two-drug combinations, it showed promise across other cancer types like melanoma and lung cancer when fed specific data. Researchers acknowledge limitations, such as the model not yet accounting for complex protein interactions with DNA or other biomolecules. Despite these constraints, the project represents a significant step toward creating digital twins that could help physicians select more effective treatments from the start, potentially reducing trial-and-error approaches in oncology.