Gene editing just got significantly more powerful — and artificial intelligence did the upgrading.
In a paper published July 16 in the journal Science, researchers at the University of California, Berkeley, report that they used AI to design entirely new CRISPR proteins that do not exist in nature — and that edit DNA more efficiently than their natural counterparts.
The work, led by Jennifer Doudna, who shared the 2020 Nobel Prize in Chemistry for her role in developing CRISPR gene editing, marks a major leap forward in the ability to customize biological tools at the molecular level.
CRISPR works like molecular scissors: a guide RNA directs a protein called a nuclease to a specific stretch of DNA, where it cuts. Scientists can then add, delete, or alter genetic information at that location. The most widely used nuclease is Cas9, which bacteria evolved millions of years ago as an immune defense against viruses. The problem with natural nucleases is that they're evolutionary products — optimized for bacteria's needs, not necessarily for human medicine or research.
That's where AI comes in. The Berkeley team focused on TnpBs — a group of tiny nucleases that are evolutionary precursors to Cas12, one of the more versatile CRISPR tools in use today. They wanted to know how far they could alter TnpB proteins while keeping them active as gene editors.
Their approach: give an AI model the final 3D structure of a TnpB protein and ask it to reverse-engineer new versions of the underlying genetic code that would produce the same folded shape — but with entirely different sequences. The AI generated thousands of candidate modifications. The team then tested which ones retained gene-editing function.
The result was a set of synthetic nucleases that not only fold correctly, but cut DNA more efficiently than the natural originals.
"Much like CRISPR democratized the ability to edit DNA at will, AI-based protein design promises to allow anyone to create totally novel properties in the protein space," said Soeren Lienkamp, a molecular biologist at the University of Zurich. He described the paper as marrying "two transformative fields": AI-guided design and RNA-guided nucleases that cut DNA and RNA strands.
The broader implication is transformational. Until now, gene editing tools were limited to what evolution happened to produce. AI protein design breaks that constraint. Scientists can now, in principle, design enzymes optimized for any specific purpose: more precise edits, fewer off-target cuts, better delivery into specific tissues, or tailored activity in particular cellular environments.
This matters enormously for medicine. CRISPR is already showing promise in clinical trials for sickle cell disease, cancer, and hereditary conditions. But natural CRISPR proteins come with trade-offs: they can be large and hard to deliver into cells, or they may cut in unintended locations. AI-designed variants could avoid those problems by design.
It also matters for agriculture. Gene-edited crops that resist drought, disease, or pests are increasingly important as climate change strains food systems. Better CRISPR tools mean safer, more precise edits in plants — and potentially faster regulatory approval.
Doudna herself described the breakthrough plainly: "Once you start tweaking things, you realize pretty quickly that while you can make changes, they ultimately produce something that isn't functional." AI has now solved that problem — designing functional proteins that evolution never found.
Not every AI-generated nuclease worked, and validating new proteins remains experimentally intensive. But the proof of concept is now published in one of the world's most rigorous scientific journals: artificial intelligence can design functional, gene-editing proteins that outperform evolution's best work. That's a capability with implications stretching far beyond the laboratory.
