Cas12Fold Accurately Predicts Cas12 Nuclease Structures to Enable Structure‐based Genome‐editing Engineering

Predicting the structurally diverse Cas12 nucleases remains challenging for general protein modeling algorithms, hindering rational engineering to enhance their genome‐editing capabilities. Here we present Cas12Fold, a deep learning framework tailored to Cas12 proteins. Cas12Fold leverages the deep evolutionary information from Cas12‐focused sequences and structures, and employs an iterative structure‐based alignment strategy to resolve conformational complexity. This approach achieves superior accuracy compared to existing methods in modeling key functional domains and capturing alternative conformations. Cas12Fold improves the structure predictions for previously refractory Cas12 proteins, including the phage‐encoded Casλ, a type V enzyme with extensive sequence and structural diversity. Accurate models generated by Cas12Fold enable robust inference of mechanistically critical residues. Guided by these predictions, structure‐based mutagenesis of DNA‐binding sites enhanced the genome‐editing efficiency of Cas12j.4. Cas12Fold thus provides a robust and generalizable platform for both mechanistic studies and the rational engineering of CRISPR–Cas12 systems.