The application discloses an
enzyme directed evolution method based on a
protein language model and multi-dimensional structure feature weighting, and belongs to the
cross field of
enzyme engineering and
bioinformatics. The method first acquires target
enzyme sequence and three-dimensional structure data, fine-tunes an ESM-1B model to identify key function anchors; after structure preprocessing, a local
interaction network is constructed through a KD-Tree and a
hydrogen bond geometry
algorithm, and candidate
mutation sites are screened by adopting multi-factor collaborative scoring; then, a sequence-activity dataset is constructed through saturation
mutation verification, closed-loop iterative optimization is performed through an ESM-2+
random forest, and site synergistic effects are mined in combination with combined mutations. The application solves the problems of complex traditional
directed evolution process, high cost, and difficulty in multi-property optimization, candidate site screening is accurate,
trial and error cost is significantly reduced, single
mutant enzyme activity can be increased by 2.0 times at most, combined
mutant enzyme activity can be increased by 2.16 times at most, the research and development cycle is greatly shortened, and the application is suitable for efficient
directed evolution of enzymes.