AI-Designed CRISPR Proteins Outperform Natural Versions
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Jennifer Doudna's team at UC Berkeley published results in Science on 16 July showing AI-designed synthetic CRISPR proteins edit the genome more efficiently than naturally occurring nucleases.
- The researchers focused on TnpB nucleases, evolutionary precursors to the commonly used Cas12, and fed an AI model the protein's final 3D conformation so it could reverse-engineer DNA template changes that preserve function.
- Doudna, who shared the 2020 Nobel Prize in Chemistry for CRISPR work, said tweaking protein sequences typically fails: "you can make changes, [but] they ultimately produce something that isn't functional."
- Independent molecular biologist Soeren Lienkamp at the University of Zurich called the paper a marriage of "two transformative fields" — AI-guided protein design and RNA-guided nucleases.
- The source notes synthetic CRISPR systems "could one day power discoveries in fields from medicine to agriculture," suggesting the AI approach could expand the gene-editing toolkit beyond the handful of naturally available nucleases like Cas9 and Cas12.
Why it matters: Conventional CRISPR is bottlenecked by a small set of naturally occurring nucleases (Cas9, Cas12) that are hard to engineer without breaking them. This study demonstrates AI can generate functional synthetic enzymes from scratch, potentially breaking the Cas9/Cas12 monopoly and opening gene-editing capabilities that evolution never produced — an advance with downstream stakes for the medicine and agriculture applications the source flags.
