The Pulse
Talus Bio Opens Ptarmigan-1 Drug Discovery Portal to Researchers
Talus Bio opened a public portal for Ptarmigan-1, its AI model for predicting small-molecule binding without requiring a protein structure. The launch offers researchers limited daily screens, while Talus’s preprint reports much larger comp
AI.info Team · Written with AI ·
On October 1, Talus Bio opened a web portal that lets researchers test Ptarmigan-1 against human protein targets, offering hands-on access to a model designed to predict small-molecule binding without first building a three-dimensional protein structure. The launch makes the company’s structure-free approach available for limited screening through an account-based service, not as an unrestricted search across billions of compounds.
What researchers can screen in the portal
Users can choose a protein and screen it against up to 10,000 compounds at a time, using Talus Bio’s library of known drugs, probes and decoys or uploading their own compounds. The company caps use at 10 screens per day. Results include predicted binding regions at residue-level resolution and possible engagement with other proteins; users can also view predictions overlaid on a structure from AlphaFold’s database.
That visual overlay does not mean Ptarmigan-1 relies on the displayed structure. Talus says the model predicts from protein sequence and two-dimensional chemical information, without constructing a three-dimensional binding pose. The company’s July description says it places protein residues and compounds in a shared mathematical space, where proximity indicates possible engagement.
Large-scale claims sit beyond the public limits
The web portal’s 10,000-compound ceiling is distinct from the much larger scale Talus reports for the underlying model. On its model page, the company says the preprint describes screening 3.4 billion compounds against 20,431 proteins, with top-ligand retrieval taking 20 H100 GPU-hours after the compound library has been embedded. Talus also advertises roughly 5,000 times the throughput of Boltz-2, but its site ties that comparison to an EGFR benchmark on one H100, measured over batched runs.
Talus’s October announcement says the model can screen 3.4 billion compounds against all human proteins in a day and reports that Ptarmigan-1 outperformed structure-based approaches in a test involving STAT6, a target with a difficult-to-model binding site. The company says the model identified new compounds for the flexible STAT6 pocket and that a third-party laboratory validated those candidates. The release does not provide the validation data or detail the experiment’s methods.
Preprint results, not a drug candidate
Ptarmigan-1’s research paper appeared on bioRxiv in July as a preprint. Its authors describe the central design this way: By decoupling molecular recognition from structure, Ptarmigan-1 recasts virtual screening as a reusable index that continuously improves as data accumulate. The paper reports tests across several target types and says the model localizes compounds to binding sites, including for targets withheld from training.
Talus says Ptarmigan draws on MARMOT, its proteomics platform for measuring protein behavior and compound interactions in living human cells. The company’s approach is intended to help investigate flexible or disordered proteins that are difficult to represent with fixed structures. That does not establish that a predicted compound will work as a medicine: Ptarmigan-1 ranks possible interactions, while experimental testing must determine whether those predictions hold up.
Access comes with a narrow daily allowance
The portal is open for account creation, but its current limits keep public use to individual targets and smaller compound libraries. Talus says larger screens and target-specific campaigns are available through direct collaboration, with expanded web and agent-based screening described as forthcoming. For now, the concrete offer is 10 screens a day, each covering no more than 10,000 compounds.