Reshaping the scissors: From directed evolution to AI-driven design for enhanced Cas performance

CRISPR/Cas gene editing holds great promise for applications in agricultural breeding and gene therapy; however, native Cas proteins face bottlenecks such as off-target effects, PAM limitations, and delivery challenges. This review explores a three-tiered, progressive, and synergistic analytical framework comprising directed evolution, rational structural design, and integrated protein engineering. Following the logic of "random exploration-precision modification-system integration," it systematically clarifies the principles, advantages, disadvantages, and complementary interactions among these three optimization techniques. By comprehensively mapping the entire technical pathway for Cas protein improvement, this study fills a gap in systematic analysis within the field and provides a theoretical framework for tiered, synergistic modification to support the development of high-performance editors. The paper further analyzes a novel research and development paradigm in which AI and protein language models profoundly empower Cas modification. It summarizes intelligent toolchains such as structure prediction, virtual mutation, and de novo design of artificial Cas proteins, and highlights the unique advantages of AI in overcoming the performance trade-offs inherent in traditional protein engineering. At the same time, it identifies existing challenges, such as data scarcity and insufficient model interpretability, and looks ahead to how multimodal gene-editing large language models and automated closed-loop experimental systems will drive the iterative development of miniaturized, broad-spectrum, and highly precise editing tools, thereby facilitating the clinical and industrial implementation of gene-editing technologies.