Integrating genetic algorithms and language models for enhanced enzyme design.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Protein Engineering

Background:

• Enzyme design is complex due to vast protein sequence space and intricate sequence-structure-function relationships.
• Large language models (LLMs) show promise for biological sequence analysis but face challenges in protein design.
• Optimizing enzymes is crucial for enabling efficient and novel chemical transformations.

Purpose of the Study:

• To develop a computational framework integrating LLMs and genetic algorithms (GAs) for enzyme optimization.
• To enhance enzyme feasibility for biochemical reactions and increase catalytic turnover rates.
• To advance the state-of-the-art in computational biocatalyst design.

Main Methods:

• Trained LLMs on extensive protein sequence data to learn residue-function-structure correlations.
• Employed GAs, guided by LLM-derived knowledge, to efficiently search for optimized enzyme sequences.
• Evaluated generated enzyme mutants across 105 biocatalytic reactions for feasibility and performance.

Main Results:

• The LLM-GA framework successfully generated enzyme mutants with improved feasibility in 90% of 105 tested biocatalytic reactions.
• In-depth analysis of seven reactions showed mutants retained structural integrity and flexibility comparable to wild-type enzymes.
• The approach demonstrated significant improvements in catalytic performance and feasibility over wild-type enzymes.

Conclusions:

• The LLM-GA framework represents a significant advancement in computational enzyme design.
• This methodology enables the creation of highly efficient biocatalysts with desirable structural properties.
• The developed approach holds potential for advancing sustainable chemical manufacturing through optimized biocatalysis.