Design of peptides with noncanonical amino acids using flow matching.

NCFlow enables the placement of any non-canonical amino acid into proteins, outperforming existing methods in structure prediction for unseen ncAAs.

• Why it matters: Expanding the amino acid vocabulary enhances protein and peptide functions, crucial for therapeutic development, but current tools are limited by scarce, biased data and inability to model diverse ncAAs.
• What they did: The team developed NCFlow, a flow matching generative model trained on small molecules and protein-ligand complexes, capable of integrating arbitrary ncAAs into protein backbones, including those never seen before.
• The result: Applying NCFlow to peptide design improved binding affinity predictions by up to -7.0 kcal/mol, demonstrating its potential to facilitate the engineering of more effective therapeutic peptides.