Enzyme design is a critical area in biocatalysis, aiming to develop efficient and selective enzymes for sustainable chemical synthesis. This review discusses the integration of molecular simulations and machine learning (ML) in enzyme design. Structure-based approaches, such as semi-rational and rational design, rely on enzyme structures to guide mutations and improve catalytic efficiency. However, these methods are limited in large-scale screening. ML techniques, enabled by big data, offer a new era for accelerated predictions and design of enzymes with desired properties. Recent advances in deep learning, such as AlphaFold2 and RoseTTAfold, have significantly improved the prediction of protein structures, enabling structure-based design. Molecular dynamics simulations and quantum mechanics/multiscale methods are also crucial for understanding enzyme mechanisms and reaction barriers. These methods, combined with ML, provide a comprehensive approach to enzyme design. The review highlights the importance of database construction and algorithm development in achieving predictive ML models. It also discusses the challenges and future perspectives of integrating traditional molecular simulations with ML for enzyme design. The application of these methods in enzyme engineering, such as improving catalytic activity, selectivity, and stability, is emphasized. The review concludes with the potential of these approaches in advancing biocatalysis for sustainable chemical production.Enzyme design is a critical area in biocatalysis, aiming to develop efficient and selective enzymes for sustainable chemical synthesis. This review discusses the integration of molecular simulations and machine learning (ML) in enzyme design. Structure-based approaches, such as semi-rational and rational design, rely on enzyme structures to guide mutations and improve catalytic efficiency. However, these methods are limited in large-scale screening. ML techniques, enabled by big data, offer a new era for accelerated predictions and design of enzymes with desired properties. Recent advances in deep learning, such as AlphaFold2 and RoseTTAfold, have significantly improved the prediction of protein structures, enabling structure-based design. Molecular dynamics simulations and quantum mechanics/multiscale methods are also crucial for understanding enzyme mechanisms and reaction barriers. These methods, combined with ML, provide a comprehensive approach to enzyme design. The review highlights the importance of database construction and algorithm development in achieving predictive ML models. It also discusses the challenges and future perspectives of integrating traditional molecular simulations with ML for enzyme design. The application of these methods in enzyme engineering, such as improving catalytic activity, selectivity, and stability, is emphasized. The review concludes with the potential of these approaches in advancing biocatalysis for sustainable chemical production.