PEGASUS: Unlocking Polarity in Cell-Permeable Cyclic Peptides Using AI Models Built on Massively Parallel Biological Assays

Cyclic peptides are a promising class of therapeutics that have the potential for oral bioavailability but are hindered by cell membrane permeability and aqueous solubility. Artificial intelligence (AI) can address the challenging multiparameter optimization of cyclic peptides, but it relies on wet lab ground truth biological data that are scarce, sparse, and dominated by hydrophobic amino acids. Here, we introduce PEGASUS, a multimodal AI model that achieves state-of-the-art performance in predicting cell membrane permeability. PEGASUS integrates an innovative high-throughput proxy biological assay (1910 PPA), which generates billions of cyclic peptides separated by permeability-related characteristics with solvent-dependent computational simulations. PEGASUS informs rules for designing cell-permeable cyclic peptides with high aqueous solubility that resemble FDA-approved therapeutics based on polarity and charge. Combining these rules with a novel generative AI, we design the first published cyclic peptides with more than two polar or ionizable fragments to achieve <i>in vitro</i> cell membrane permeability.