The abstract is read at the publisher; the summary is JClub's.
FoldMapper designs protein sequences that recover target backbones with 60.0% best-of-100 recovery on TED domains and 64.9% on PDB, while improving structure self-consistency.
- Why it matters: Inverse folding typically needs evolutionary profiles from an alignment built using the query sequence, creating a circular dependency. FoldMapper aims to use evolutionary statistics without requiring a sequence homology search.
- What they did: It retrieves structural homologues using shape alone, aggregates amino-acid identities into per-residue SPSF profiles, and encodes them with a multiscale graph encoder for structure-conditioned message passing.
- The result: It achieves 51.3%/53.1% on PPIBindBench non-peptide/peptide binders and sits on the recovery–diversity Pareto front; structure self-consistency surpasses baselines on RFdiffusion de novo monomer and binder backbones.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.