From traditional to AI-driven: The evolution of intelligent enzyme engineering for biocatalysis

Enzyme engineering involves enhancing enzyme function and application through multidimensional technological systems. Its development encompasses elucidating sequence-structure-function relationships, exploring fitness landscapes, and multiscale regulation. Conventional enzyme engineering strategies include directed evolution (DE), rational/semi-rational design, residue co-evolution, and de novo design. DE mimics natural selection but is limited by high-throughput screening efficiency. Rational/semi-rational design integrates computational simulation with experimental validation to regulate enzyme performance. Residue co-evolution combines sequence co-evolution analysis and kinetic simulations, and de novo design is dedicated to achieving precise protein folding through physical modeling. In recent years, the breakthrough progress in machine learning (ML), especially deep learning (DL), has significantly enhanced the efficiency of all the above-mentioned methods; by accurately predicting mutational effects and efficiently exploring discontinuous sequence space, it provides powerful tools to improve or supplement the above strategies, thereby aiding in escaping local fitness optima traps. This review summarizes common strategies in enzyme engineering, including directed evolution, rational/semi-rational design, residue co-evolution, and de novo design, and further introduces the latest applications of ML and DL models in each of these fields. Although challenges persist, such as force field accuracy limitations, mutation sampling constraints, experimental throughput limitations, and epistatic effects, more comprehensive multimodal foundation models in the future are expected to integrate cross-scale parameters for intelligent design, and the establishment of standardized enzymology databases will enhance prediction reliability. Overall, AI-empowered enzyme engineering will drive a profound transformation toward a predictable and highly efficient pathway for enzyme design, providing more precise and powerful solutions for biocatalysis.