AI-designed CRISPR enzymes improve genome editing

Scientists have used AI models to design synthetic CRISPR proteins that edit genomes more efficiently than naturally occurring versions. The results were published on 16 July in Science and could eventually support research in medicine, agriculture and other fields that rely on precise genetic changes.

CRISPR depends on RNA-guided nucleases, enzymes that use a guide RNA to find a target DNA sequence and cut it like molecular scissors. Common systems such as Cas9 and Cas12 were adapted from bacterial immune machinery, but altering these enzymes is difficult because small changes can disrupt the sequence of steps needed for gene editing to work.

Jennifer Doudna and collaborators focused on TnpBs, tiny nucleases that are evolutionary precursors to Cas12. The team gave an AI model the final conformation of a TnpB protein and asked it to reverse-engineer DNA-template changes that would preserve that shape. The approach generated thousands of potential protein designs, offering a way to explore functional nucleases far faster than conventional trial-and-error experiments.