CatIF-RL: Activity-Oriented Enzyme Sequence Design by Steered Inverse Protein Folding

Area of Science:

• Computational Biology
• Protein Engineering
• Enzyme Design

Background:

• Protein inverse folding models generate sequences for given structures but lack functional optimization.
• Enzyme engineering requires designing variants with improved catalytic activity.

Purpose of the Study:

• To develop a framework, CatIF-RL, that steers inverse folding models toward designing enzyme variants with enhanced predicted catalytic activity.
• To enable functional specialization of protein design models.

Main Methods:

• Adapted a graph-based denoising diffusion inverse folding model to enzyme structural data.
• Introduced activity-oriented preference signals using predicted catalytic constant (kcat) as the optimization objective.
• Employed generative dataset curation and group-relative policy optimization (GRPO) for specialization.

Main Results:

• CatIF-RL achieved an approximately 4-fold increase in predicted kcat compared to native enzymes.
• Maintained sequence recovery (0.55) and predicted structural quality comparable to existing methods.
• Demonstrated support for motif-preserving partial sequence design.

Conclusions:

• CatIF-RL provides a practical framework for activity-oriented enzyme design.
• Offers a generalizable strategy for steering structure-conditioned protein generation towards functional optimization.