Leveraging Auxiliary Potentials in RFDiffusion for the Design of NA-Binding Proteins.

Area of Science:

• Computational biology
• Protein design
• Nucleic acid-protein interactions

Background:

• Designing de novo nucleic acid-binding proteins is complex due to target size and electrostatics.
• Existing platforms like RFDiffusion require refined strategies for nucleic acid target design.
• Two main design paradigms, de novo generation and motif-scaffolding, are explored.

Purpose of the Study:

• To systematically investigate and compare de novo generation and motif-scaffolding strategies for nucleic acid-binding protein design.
• To evaluate enhancements to standard design frameworks, including rigid-body-averaged and px0-guided potentials.
• To assess the impact of these strategies on interface quality, foldability, and target specificity.

Main Methods:

• Systematic comparison of de novo generation and motif-scaffolding across 11 diverse nucleic acid targets.
• Evaluation of rigid-body-averaged potential and px0-guided potential enhancements.
• Assessment of protein foldability using Boltz2 and computational metrics.

Main Results:

• The px0-guided potential yields favorable interfaces but compromises foldability.
• Rigid-body averaging alone improves refolding coherence without direct interface gains.
• Combining px0 and averaging strategies balances interface metrics and computational foldability.
• Motif-scaffolding performance is context-dependent, with auxiliary potentials showing variable success.
• Newer models may produce superficially good metrics but lack specificity and diversity.

Conclusions:

• Modular, tunable potentials, especially when combined, offer advantages over monolithic approaches for nucleic acid-binding protein design.
• The study provides a practical comparative assessment to guide future de novo design efforts.
• Balancing interface optimization with foldability is crucial for successful nucleic acid-binding protein design.