EnsembleCycPerm model accurately predicts cyclic peptide permeability by integrating solvent-dependent conformational ensembles

Cyclic peptides are a promising therapeutic modality, particularly for challenging targets like protein-protein interactions. However, their clinical utility is often severely limited by inadequate membrane permeability, restricting both intracellular target access and oral bioavailability. This poor permeability arises from complex, solvent-dependent conformational ensembles, making their behavior difficult to predict. Current drug discovery efforts face significant hurdles due to the inherent structural complexity and laborious synthetic demands of these macrocycles, necessitating better predictive tools to overcome the permeability bottleneck.

Researchers developed EnsembleCycPerm, a novel modeling framework inspired by molecular chameleons, designed to predict cyclic peptide permeability. The model integrates sequence and structural information by explicitly representing solvent-dependent conformational ensembles in both aqueous and nonpolar environments. Validation was performed using the CycPeptMPDB dataset, evaluating performance under both random split and scaffold split methodologies. The study also analyzed one-residue analog-pairs within the testing set to assess the model's ability to capture experimentally meaningful trends.

The EnsembleCycPerm model demonstrated strong predictive performance for cyclic peptide permeability. Under a random split evaluation, it achieved an MAE of 0.29, a Pearson R of 0.85, and an R2 of 0.70 on the CycPeptMPDB dataset. Performance remained robust under a more challenging scaffold split evaluation, yielding an MAE of 0.34 and a Pearson R of 0.68. Beyond overall prediction, the model successfully captured experimentally meaningful ΔPAMPA trends for one-residue analog-pairs in the testing set, supporting its utility for lead optimization. Model interpretation highlighted ΔPSA3D (polarity shielding) as a key mechanistically interpretable feature strongly associated with permeability. These findings suggest that permeability emerges from ensemble reorganization rather than a single conformational transition.

The model achieved strong predictive performance with an MAE of 0.29, R of 0.85, and R2 of 0.70 on the CycPeptMPDB under random split evaluation.