Improving Inverse Folding for Peptide Design with Diversity-regularized Direct Preference Optimization

Inverse folding models play an important role in structure-based design by predicting amino acid sequences that fold into desired reference structures. Models like ProteinMPNN, a message-passing encoder-decoder model, are trained to reliably produce new sequences from a reference structure. However, when applied to peptides, these models are prone to generating repetitive sequences that do not fold into the reference structure. To address this, we fine-tune ProteinMPNN to produce diverse and structurally consistent peptide sequences via Direct Preference Optimization (DPO). We derive two enhancements to DPO: online diversity regularization and domain-specific priors. Additionally, we develop a new understanding on improving diversity in decoder models. When conditioned on Open-Fold generated structures, our fine-tuned models achieve state-of-the-art structural similarity scores, improving base ProteinMPNN by at least 8%. Compared to standard DPO, our regularized method achieves up to 20% higher sequence diversity with no loss in structural similarity score. Engineering biopolymers that fold into desired 3D structures, a computational challenge known as inverse protein folding problem, has broad applications in drug discovery and material science (Yang et al., 2023; Dill et al., 2008; Abascal & Regan, 2018). Several approaches for inverse folding have been adopted over the past decades, from molecular dynamics simulations to machine learning approaches (Dauparas et al., 2022b; Shanker et al., 2023; Hsu et al., 2022a; Yi et al., 2023; Correa, 1990).