CasMiner: a deep learning tool for high-throughput mining and rational design of efficient Cas9

Abstract Since its inception, the CRISPR-Cas system, particularly Cas9, has demonstrated immense potential for life science applications, but expansion of the Cas9 toolkit is constrained by sequence alignment-based strategies for mining and optimization. Here, we developed CasMiner, a deep learning model for discovering and engineering novel Cas9 proteins. CasMiner achieved 99.63% accuracy in predicting Cas9s, and identified VpCas9 from public databases. Experimental validation showed that VpCas9 exhibits robust double-strand cleavage activity. Combining CasMiner and evolutionary analysis, we engineered three mutants with markedly increased structural rigidity and positive charge. In vivo cleavage assays revealed that the mutant VPM2-3 achieved a higher average editing efficiency in rice callus and maize protoplasts than the wild-type VpCas9, whose editing efficiency rivals that of SpCas9. This study thus establishes a comprehensive platform for mining and engineering Cas9 proteins, and provides VpCas9 and derivative nucleases as powerful tools that greatly broaden the horizon for genome editing applications.