We’re excited to release a major update to the Boltz repo: v0.3.0. This release contains several important new features, including our confidence model and memory-efficient inference. Give it a try! https://github.com/jwohlwend/boltz
Comments
Log in with your Bluesky account to leave a comment
The confidence model provides metrics such as pLLDT, pTM, PAE, and automatically ranks predictions based on an aggregate metric. We also developed a low-memory mode that kicks in automatically passed 512 tokens, allowing users to predict the structure of larger complexes, even on smaller GPUs!
Awesome! I'll read up and give it a go. Any plans to implement additional experimental constraints, e.g. C alpha distances or cross-linking data (maybe already possible?).
That will be super useful for small molecule compounds. Can it be extended? Constraining distances between protein chains alone is also very useful (eg when experimental distance constraints are known for a challenging complex).
Pocket conditioning will also be supported for polymer chains. It allows to specify any residue(s) that is known to be in contact with another chain/molecule
Comments