Reproduce “Inference-Time Projection for Physically Valid Biomolecular Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Qurat-ul-ain, Yee Whye Teh, Charlotte M. Deane, Matteo Cagiada
The method adds bounded coordinate corrections during biomolecular diffusion sampling, separating inter-chain clashes and repairing ligand geometry in the final 40 of 200 steps without changing model weights. Later predictions show reduced penetration. On 52 CASP16 targets, Boltz-2 best-of-five physical validity rises from 0.538 ± 0.069 to 1.000 ± 0.000, while LDDT is unchanged, 0.453 ± 0.044 versus 0.452 ± 0.044. This empirical target-level result does not guarantee every sample or physical property is valid.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.