APCyc Generates Target-Aware Cyclic Peptides with Bayesian Property Guidance

APCyc uses an expanded residue vocabulary encoding cyclization sites and linkage types to generate cyclic peptides de novo, with Bayesian posterior guidance jointly optimizing multiple physicochemical drug properties. It addresses a gap in generative models trained primarily on linear peptide data that miss cyclization-specific structural constraints.