Doudna Lab Designs Novel Gene Editors Using Meta AI
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Jennifer Doudna's team published in Science a new platform that uses Meta's inverse protein model to design never-before-seen TnpB nucleases active in human, plant, and bacterial cells — the lab's first AI-designed proteins rather than AI-discovered ones.
- Petr Skopintsev fed TnpB into Meta's inverse-folding model, which inverts AlphaFold by taking a backbone structure as input and outputting sequences that fold into it, then restricted the model's output to known DNA-binding sites to generate candidate enzymes.
- Isabel Esain-Garcia validated the AI-generated proteins in genome editing experiments and said the tiny size of TnpBs makes them much easier to pack for organ-specific delivery — useful for personalized medicine and climate-adapted crops.
- Outside expert Benjamin Kleinstiver (Mass General Brigham / Harvard) called the method's IP implications the most interesting near-term angle, noting that sequences diversified past patent similarity thresholds 'may be able to create IP-evading enzymes' in a field already defined by broad patent fights.
- David Baker has been the chief figure in AI protein design; Doudna told Fierce her team works closely with Baker's, calling it a '1 + 1 = 3' collaboration, and declined to comment on whether a new startup is in the works beyond her existing Mammoth Biosciences.
Why it matters: Kleinstiver flagged the IP angle as the most interesting near-term consequence: the platform's ability to diversify sequences past patent similarity thresholds gives labs a route around existing gene editing IP — a major area of contention, as the recent Prime vs. Beam patent fight shows. Doudna's existing startup Mammoth already specializes in delivering small editors like these TnpBs.
