AI model targets previously undruggable proteins in living cells

Talus Bio's Ptarmigan-1 predicts binding sites for unstructured proteins without requiring 3D structures

Talus Bio recently released Ptarmigan-1, an AI model that predicts small molecule binding sites for unstructured proteins within native cellular environments. The model operates without needing defined protein 3D structures—a key advance for targeting proteins previously considered undruggable. It has demonstrated superiority over structure-based methods in a STAT6 inhibitor series and targets approximately 40% of human proteins lacking stable 3D structures, including the brachyury protein (TBXT gene) for chordoma treatment. The system uses label-free proteomics to measure protein interactions in living cells without removal or tagging. Talus Bio has generated at least five years of proprietary cellular protein data. However, the model's predictive accuracy for unstructured proteins remains unverified beyond the STAT6 proof-of-concept study, and it has not been tested on the specific target in that study. This represents a significant early-stage capability for drug discovery but requires further validation before clinical application.

Source: genengnews.com