Protein Inverse Folding From Structure Feedback

2025-06-03

AI Summary

This work addresses the protein inverse folding problem—designing amino acid sequences that optimally adopt a given target 3D structure. We propose a novel closed-loop optimization framework, introducing Direct Preference Optimization (DPO) to this task for the first time. The framework iteratively executes four stages: sequence sampling, structural evaluation via RoseTTAFold/AlphaFold2 variants to generate structure-prediction labels, preference feedback derived from structural similarity (e.g., TM-Score), and end-to-end sequence optimization guided by DPO. On the CATH 4.2 benchmark, our method achieves a mean TM-Score of 0.81—outperforming baselines by +0.04. For hard-to-fold targets, multi-iteration refinement yields a relative TM-Score improvement of 79.5%, demonstrating substantially enhanced modeling capability for complex conformations. This represents the first application of preference-based learning to inverse folding and establishes a principled, feedback-driven paradigm for structure-aware sequence design.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Humans and AI: Learning Human Values and Preferences
• Machine Learning: Learning Preferences or Rankings

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

The inverse folding problem, aiming to design amino acid sequences that fold into desired three-dimensional structures, is pivotal for various biotechnological applications. Here, we introduce a novel approach leveraging Direct Preference Optimization (DPO) to fine-tune an inverse folding model using feedback from a protein folding model. Given a target protein structure, we begin by sampling candidate sequences from the inverse-folding model, then predict the three-dimensional structure of each sequence with the folding model to generate pairwise structural-preference labels. These labels are used to fine-tune the inverse-folding model under the DPO objective. Our results on the CATH 4.2 test set demonstrate that DPO fine-tuning not only improves sequence recovery of baseline models but also leads to a significant improvement in average TM-Score from 0.77 to 0.81, indicating enhanced structure similarity. Furthermore, iterative application of our DPO-based method on challenging protein structures yields substantial gains, with an average TM-Score increase of 79.5% with regard to the baseline model. This work establishes a promising direction for enhancing protein sequence design ability from structure feedback by effectively utilizing preference optimization.

Problem

• Design amino acid sequences for desired 3D protein structures
• Optimize inverse folding using feedback from folding models
• Improve sequence recovery and structure similarity metrics

Innovation

• DPO fine-tunes inverse-folding model
• Uses folding model feedback
• Iterative method improves TM-Score