AI Designs Functional CRISPR-Like Nucleases That Work in Cells

Overview

• Researchers published a Science paper on July 16, 2026 that used an AI-guided, structure-informed workflow to design synthetic versions of the minimal TnpB nuclease family called SynTnpBs.
• The team combined the ESM Inverse Folding (ESM-IF1) model with residue constraints drawn from evolution to generate protein sequences that diverge substantially from natural TnpB variants.
• Laboratory screens first in bacteria and then in plant and human cells showed many SynTnpBs retained or surpassed the activity of natural TnpB enzymes.
• Cryo-electron microscopy structures of the most divergent SynTnpBs revealed new electrostatic and hydrogen-bond networks that stabilize the RNA–DNA interface and confirm the AI designs adopt functional conformations.
• The work establishes a proof of concept that structure-guided AI design can produce active, non-natural RNA-guided nucleases and points to potential uses in research and agriculture while noting that safety, broader functionality, and regulatory steps remain to be explored.