EnerBridge-DPO: Energy-Aware Markov Bridge Inverse Folding for Protein Sequence Design

Area of Science:

• Computational Biology
• Protein Engineering
• Bioinformatics

Background:

• Designing protein sequences with favorable energetic properties is a key challenge in inverse folding.
• Current deep learning methods often prioritize sequence recovery over energy-based preferences.

Purpose of the Study:

• To develop an energy-aware inverse folding framework for protein complex design.
• To integrate Markov bridge sequence generation with preference optimization for improved energetic properties.

Main Methods:

• Proposed EnerBridge-DPO framework integrating Markov bridge sequence generation and preference optimization.
• Introduced a Bridge-DPO objective using energy-related preference pairs to guide sequence generation.
• Incorporated an energy-constrained loss based on mutation-induced binding free-energy changes (ΔΔG).

Main Results:

• EnerBridge-DPO maintained competitive inverse-folding performance.
• Achieved lower predicted energy scores for protein complexes under selected computational scoring functions.
• Demonstrated competitive ΔΔG prediction performance on SKEMPI, with minor, statistically inconclusive gains in overall metrics.

Conclusions:

• Integrating energy-related preferences into Markov bridge inverse folding can enhance computationally predicted energetic profiles.
• Further experimental validation is needed to confirm the thermodynamic stability of designed proteins.
• EnerBridge-DPO shows promise for designing protein complexes with improved energetic characteristics.