bioRxivJClub
Biomolecular design through latent reasoning with RFD4-Proteina
bioRxiv · · Preprint · Open access
Corley, Didi + 43 more
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
RFD4-Proteina improves in silico protein design success rates at higher generation throughput than all previous evaluated methods, using intrinsic and extrinsic reasoning in a shared chemical–conformational latent space.
- Why it matters: Biomolecular design is constrained by limited molecular landscapes and inefficient exploration, limiting how well deep learning models can search for chemically and geometrically compatible interacting structures.
- What they did: The model learns a latent space spanning standard and modified amino acids, sugars, DNA, and RNA, with force-field supervision for physically plausible atomic reconstructions, and uses recurrent refinement plus constraint-guided search.
- The result: It achieves higher success rates across benchmarks and matches or exceeds specialist methods for side-chain packing, sequence design, conformational sampling, and refinement, and is experimentally validated with binders, phosphorylation/glycosylation control, and de novo enzyme design for active pharmaceutical ingredient synthesis.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.
- Open access
Free. We’ll email you when a new paper cites this one or any other paper you follow on JClub — at most one email a day. Every email has a one-tap unsubscribe.