This study presents a computational approach for designing target-specific peptide inhibitors by integrating a Gated Recurrent Unit-based Variational Autoencoder (GRU-VAE) with Rosetta FlexPepDock for peptide sequence generation and binding affinity assessment, followed by molecular dynamics (MD) simulations for refinement. Applying this to β-catenin and NF-κB essential modulator (NEMO), six out of twelve β-catenin inhibitors showed improved binding affinity (best with a 15-fold improvement), and two out of four NEMO inhibitors showed enhanced binding. This highlights the successful integration of deep learning and structure-based modeling for peptide design.
