Deep learning–driven prediction of on-target activity, off-target risk, and repair outcomes in CRISPR/Cas9: current landscape and multi-scale perspectives

The CRISPR/Cas9 system has emerged as a transformative tool in genome editing, playing a pivotal role in enabling precise genetic engineering. Achieving high on-target efficiency while minimizing off-target activity is critical for translating CRISPR/Cas9 into reliable experimental and therapeutic applications. Conventional off-target detection methods are labor-intensive and cost-prohibitive, limiting their scalability. The integration of artificial intelligence has markedly reduced detection costs and substantially increased throughput. Early shallow learning models in the CRISPR/Cas9 domain, although effective in basic classification tasks, exhibited limited feature representation and poor generalization. With advances in algorithms and computational power, deep learning architectures have significantly improved off-target prediction accuracy. However, a critical blind spot remains, most current models operate predominantly at the sequence level, overlooking the downstream functional consequences of genome edits. This review summarizes the current landscape of AI-driven CRISPR/Cas9 prediction methods and proposes a forward-looking “three-layer framework” that integrates molecular, cellular, and tissue dimensions. By linking nucleotide-level edits to protein alterations, cellular functional changes, and tissue-specific responses, this framework aims to bridge the gap between sequence-based predictions and phenotypic outcomes, thereby advancing the precision and translational potential of CRISPR/Cas9 technologies.