- Anthropic says its Claude AI found a new gene system in viruses during a 21-hour search, beside repeating DNA like CRISPR's.
- Anthropic doesn't yet know what Claude's CRISPR-like find does, but says similar known systems can cut, copy and paste DNA.
- Anthropic says earlier studies already found the enzyme in Claude's find, but Claude appears to be first to spot the repeating DNA beside it.
- Feng Zhang, a CRISPR pioneer, called Anthropic's Claude find "genuinely intriguing," but some scientists told Wired it doesn't look like a new CRISPR.
- Anthropic's paper on the CRISPR-like Claude find hasn't been through formal peer review, and Anthropic says its work to learn what the system does goes on.
Anthropic says its Claude AI found a new set of genes in viruses that infect bacteria. The genes sit beside a long row of repeating DNA that looks like part of CRISPR, the Nobel Prize-winning gene-editing tool. Anthropic doesn't yet know what the system, named ART, does. But it says the few known systems like ART can all cut, copy and paste DNA.
Anthropic Claude enzyme discovery announcement
Anthropic shared the result on September 23 and says its scientists gave Claude only broad directions. They asked Claude to search huge DNA databases for enzymes called reverse transcriptases. These proteins copy RNA, a molecule cells make from DNA, into DNA. Anthropic's paper says such proteins are already used in biotech, including gene editing.
Over about 21 hours, about 950 Claude sessions gathered about 200,000 groups of related enzymes. The search ended with 19 reports on possible finds, and most of those leads didn't hold up. One Claude session had noticed a long row of repeating DNA beside the gene for an unusual enzyme. It counted the repeats, searched earlier research for the pattern and filed a report for Anthropic's scientists.
Anthropic says earlier studies already found that enzyme, but Claude appears to be first to notice the repeats and a partner gene beside it. Wired says one was a 2021 paper by Texas A&M microbiologist Jason Gill and colleagues. Seth Shipman of the Gladstone Institutes told Wired he doesn't think ART is a new CRISPR system. "The novel thing is how they found it, not what it is," he says, though he still calls the discovery interesting.
Wired report on scientists' reactions
Anthropic's own paper says ART's repeats differ from CRISPR's, and none of the ART systems found comes with the genes CRISPR usually has. The paper says ART's enzyme is a close relative of the enzymes in retrons, another system bacteria use to fight viruses. Anthropic's lab tests suggest ART's row of repeats is turned into several short pieces of RNA. The paper hasn't been through formal peer review.
Anthropic preprint on ART enzymes
Feng Zhang, an MIT professor and CRISPR pioneer, reviewed Anthropic's paper and called the finding "genuinely intriguing." Fyodor Urnov of UC Berkeley praised Anthropic in Wired for sharing its discovery with the world. But Stanford professor Le Cong told Wired, "The experiments are still in the queue. The PR is already live." Cong also credits Anthropic's scientists for picking a problem Claude is good at. It's still unknown whether ART could be used to edit genes.
