RoseTTAFold diffusion-guided short peptide design: a case study of binders against Keap1/Nrf2.

In this study, we proposed a novel comprehensive computational framework that combines deep generative modeling with in silico peptide optimization to expedite the discovery of bioactive compounds. Our methodology utilizes RFdiffusion, a variation of the RoseTTAFold model for protein design, in tandem with ProteinMPNN, a deep neural network for protein sequence optimization, to provide short candidate peptides for targeted binding interactions. As a proof-of-concept, we focused on Keap1 (Kelch-like ECH-associated protein 1), a key regulator in the Keap1/Nrf2 antioxidant pathway. To achieve this, we designed peptide sequences that would interact with specific binding subpockets within its Kelch domain. We integrated machine learning models to forecast essential peptide properties, including toxicity, stability, and allergenicity, thus enhancing the selection of prospective candidates. Our in silico screening identified eight top candidates that exhibited strong binding affinity and good biophysical characteristics. The candidates underwent additional validation via comprehensive molecular dynamics simulations, which confirmed their strong binding contacts and structural stability over time. This integrated framework offers a scalable and adaptable platform for the rapid design of therapeutic peptides, merging breakthrough computational techniques with focused case studies. Furthermore, our modular methodology facilitates its straightforward adaptation to alternative protein targets, hence considerably enhancing its potential influence in drug development and discovery.