Talus Bio launches Ptarmigan-1
AI-native pharmaceutical company Talus Bio has launched Ptarmigan-1, billed as the first AI model that can predict small-molecule binding across the entire human proteome without needing three-dimensional structures, and has opened public access.
Key points
- Skipping structure and reading sequence directly: Traditional AI drug discovery relies on 3D modeling of small molecules fitting into protein pockets, but about 40% of human proteins, including many disease-related transcription factors and regulatory proteins, are too flexible and have no fixed structure to grab onto. Ptarmigan-1 simply drops the structure-modeling step.
- Outperforms traditional methods in tests: In an evaluation on a set of publicly disclosed STAT6 inhibitors (a validated target for inflammatory disease whose binding site is disordered and hard to model), Ptarmigan-1 outperformed structure-based methods in candidate screening, even though it had never seen the target or the molecules during training; the novel small molecules it discovered have been validated by a third-party laboratory.
- Complementary rather than a replacement: The company's CEO said that after the 2024 Nobel Prize in Chemistry honored structure prediction, the field has increasingly focused on "using structure to drive discovery"; Ptarmigan-1 is a complementary route for the other half of proteins whose structures cannot be computed.
Source: read the original. This text is a summary for reference only.
