HighPlay3: Structure-Integrated de novo cyclic peptide design balancing binding affinity and membrane permeability

Cyclic peptides represent a highly promising class of next-generation therapeutics owing to their favorable target-binding affinity and selectivity; however, improving membrane permeability while preserving target-binding activity remains a major challenge that limits their oral bioavailability and clinical translation. Although artificial intelligence has accelerated cyclic peptide discovery, existing computational frameworks predominantly focus on optimizing binding affinity while neglecting membrane permeability. To address this gap, we developed HighPlay3, a structure-integrated de novo design platform that uses iterative structural evaluation and membrane permeability prediction to generate cyclic peptides with balanced target-binding affinity and membrane permeability. HighPlay3 was applied to two representative targets, MDM2 and GABARAP, followed by a multistep screening cascade and experimental validation. Remarkably, 80% and 60% of the designed cyclic peptides targeting MDM2 and GABARAP, respectively, exhibited micromolar binding affinities. Further Caco-2 permeability evaluation of the SPR-active GABARAP-targeting candidates identified GAB-D4 as having the highest apparent permeability among those tested, with an efflux ratio below 1, consistent with no apparent net efflux under the assay conditions. Together, these results support HighPlay3 as a feasible framework for the early-stage design and screening of cyclic peptide binders with balanced target-binding affinity and membrane permeability.