AI-assisted transformation of PD-L1 inhibitory peptides into small molecules using amino acid mapping descriptor and activity improvement through structure-based drug design.

We present a novel strategy for the conversion of macrocyclic peptides into small molecules to identify potent and membrane-permeable protein-protein interaction (PPI) inhibitors. Our approach utilizes amino acid mapping (AAM) descriptors in conjunction with a generative artificial intelligence (AI) structure generator, enabling the design of small molecules that capture essential interaction profiles without relying on the original peptide scaffold. Using this method, we successfully transformed the known PD-L1 inhibitor Peptide-71 (Ki = 0.004 μM) into novel small-molecule inhibitors. Notably, compound 1, derived from the key interaction sites of Peptide-71, retained critical binding features and achieved a two-fold increase in a measure of activity per molecular weight, i.e., binding efficiency index (BEI: 9.2 vs. 4.7), despite a ∼ 10,000-fold decrease in absolute Ki (39 μM vs. 0.004 μM). Furthermore, structure-based optimization utilizing protein structural information led to the discovery of compound 2c, which exhibited a significant increase in potency (Ki = 5.7 μM), representing a more than six-fold increase over 1. These results demonstrate the potential of the combined AI-AAM and structure-based optimization strategy to efficiently convert peptides into small-molecule PPI inhibitors and accelerate the discovery of novel therapeutics.