Designing Beyond Nature
Unlike previous AI-driven approaches that tend to produce sequences closely resembling their natural templates, the Doudna lab’s strategy generates proteins with substantially different sequences while maintaining or improving enzymatic function. The team screened these AI-designed proteins for activity in bacterial, plant, and human cells, finding that many retained or exceeded the editing efficiency of the wild-type enzyme. Ssciencemediacentre
Eeurekalert
Cryo-electron microscopy revealed that the most divergent engineered variants formed previously undescribed stabilizing interactions at the RNA-DNA interface, including a novel TAM-bound conformation not seen in natural enzymes. These represent the first experimentally determined structures of AI-designed RNA-guided nucleases. Eeurekalert
Ssciencemediacentre
Balancing Activity and Specificity
Independent experts noted the work’s rigor while flagging nuances. José Luis Villanueva, commenting for the Science Media Centre Spain, observed that some of the most active variants also showed higher off-target editing, “reflecting the importance of balancing or incorporating these other factors into the design or evaluation process”. Marc Güell of Pompeu Fabra University called the study “another step forward in the growing convergence between AI and synthetic biology,” noting that while design is increasingly computational, it “still relies on experimentation”. Ssciencemediacentre
Expanding the Toolbox
The practical upshot, according to structural biologist Sergi Rodà, is that the work “democratizes the protocol to design your own new-to-nature RNA-guided nucleases”. The platform could accelerate development of editing tools tailored for genetic disease treatment and crop engineering — applications where compact, high-specificity nucleases are in demand. The paper’s preprint had been posted on bioRxiv in December 2025, with the peer-reviewed version now appearing in print. Oorcid
Bbiorxiv
Ssciencemediacentre
