HFGuidedDesign: de novo design of cyclic peptide binders via structure-guided discrete diffusion

cyclic peptide design framework that integrates a discrete diffusion model with external structure guidance. By incorporating the high-accuracy complex structure predictor HighFold, the framework performs real-time structural evaluation during reverse diffusion sampling and dynamically steers sequence generation toward cyclic peptides with favorable structural plausibility and binding potential. The discrete diffusion model is trained using a two-stage strategy, including pre-training on peptide monomers and fine-tuning on peptide-protein complex structures. In design tasks targeting two distinct proteins, we evaluate two classical cyclization strategies-head-to-tail and disulfide bond cyclization. The resulting cyclic peptides achieved sequence design success rates of 75% and 66.7% for the two targets, demonstrating the effectiveness and generalizability of the framework. This study establishes an innovative and scalable computational framework for sequence-based cyclic peptide design, facilitating the development of peptide-based ligands for protein targeting.