The abstract is read at the publisher; the summary is JClub's.
NCFlow is a flow-matching generative model that can place arbitrary noncanonical amino acids into a protein backbone using only atom types and bond connectivity, generalizing beyond training.
- Why it matters: Expanding peptide and protein chemical space beyond the 20 canonical amino acids is needed to engineer therapeutic peptides, especially macrocycles, but current design tools struggle with ncAAs and data are scarce and biased.
- What they did: NCFlow is pretrained on millions of small-molecule structures and protein–ligand complexes, then fine-tuned on native noncanonicals from proteins, and it outperforms AlphaFold3-based methods for unseen ncAAs.
- The result: In four protein–peptide complex test cases, a peptide design pipeline with a hybrid deep-learning and alchemical molecular-dynamics scoring strategy finds variants with predicted binding affinity improved by up to -7.0 kcal/mol.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.