An AI-designed gene editor now works in human cells and rice plants

An AI-generated protein has become a functioning tool for rewriting DNA. OpenCRISPR-1 has edited cultured human cells and, after adaptation, rice plants. Those experiments give researchers something they can use, measure and redesign: a new biological instrument created with a protein language model.[1][3]

Designing the tool that makes the edit

CRISPR editing combines a protein that acts on DNA with a guide RNA that directs it towards a chosen sequence. The guide provides the address. The protein performs the molecular work. Researchers can use this system to investigate a gene by changing it and observing what happens in a cell or organism.[5]

Much of the familiar CRISPR toolkit grew from bacterial systems. Profluent’s July 2025 Nature paper described language models trained using more than a million CRISPR operons—groups of associated genes. Some generated editors retained useful activity despite being hundreds of amino-acid changes from the standard SpCas9 protein.[1]

A protein language model learns relationships among the building blocks of proteins. Here, its output becomes a candidate molecular machine. That makes the experimental handover particularly revealing: a sequence must produce a protein that folds and acts in a cell. The successful designs extend the set of tools available for biological research.[1]

A working editor in two different settings

An independent study published in Genome Medicine in May 2026 compared OpenCRISPR-1 with Cas9 across 28 DNA targets in HEK293T human cells. Average editing frequencies were similar. The researchers also found fewer unintended edits at the off-target locations they investigated and converted OpenCRISPR-1 into prime editors, which combine DNA targeting with machinery that writes a specified change.[2]

Performance depends on the biological setting and test. The 2026 team found somewhat lower activity in a human lung-derived cell line. A separate 2025 comparison by Tian and colleagues also reported less favourable OpenCRISPR-1 results, while finding strong performance from a naturally derived editor called FrCas9. These results support testing the tool for its intended job rather than assigning it one universal accuracy score.[2][4]

The rice experiment shows how a design travels

In New Phytologist, researchers adapted the editor for rice and demonstrated gene disruption, base editing and prime editing. They also regenerated edited plants. This matters because a protein that functions in one laboratory cell line does not automatically become a useful plant-science tool. The rice work established another setting in which researchers could put the design to work.[3]

The team made several of its research constructs available through Addgene, including the rice-adapted editor and versions for changing individual DNA bases. Sharing physical research materials complements publishing the sequence: another laboratory can investigate an identifiable tool and report how it behaves.[7]

A new starting point for biological engineering

Profluent’s documentation records the original public release on 22 April 2024, preceding the Nature paper. It provides the editor’s sequence and a compatible guide RNA, with commercial use subject to its licence terms. The later studies show what can happen after such a release: other groups adapt the design, compare it with alternatives and develop additional editing functions.[6]

Genome editing is also a discovery technique. Changing a gene helps scientists connect a DNA sequence with a biological effect, which is why these tools are valuable across basic research as well as medicine. A larger choice of working editors can therefore benefit experiments long before any one design becomes a treatment.[5]

The achievement is a reusable piece of molecular engineering. AI generated a candidate protein; experiments established that it could perform a useful task; further laboratories began turning it into tools for their own questions. Each successful adaptation makes that initial design more useful.

Report an error or suggest a correction