An operational perturbation proteomics-based virtual cell model.

ProteinTalks, a virtual cell model trained on over 38 million proteomic measurements, accurately predicts drug responses and resistance in breast cancer models.

• Why it matters: Understanding protein dynamics in response to perturbations is crucial for improving drug discovery and personalized cancer treatments, yet existing models lack large-scale, time-resolved data and interpretability.
• What they did: The team generated extensive temporal proteomic data from breast cancer cell lines and developed ProteinTalks, a model that learns transferable dynamic representations through a pretraining framework.
• The result: ProteinTalks outperforms benchmarks in predicting drug efficacy, synergy, and resistance, and extends its predictive power to patient-derived organoids and biopsies, advancing in silico drug discovery.