The pioneers of the CRISPR gene-editing tool won a Nobel prize in 2020, and for good reason. The breakthrough has revolutionised biology, and is also transforming medicine and farming. Now, AI company Anthropic has announced that it has discovered an enzyme system with properties “reminiscent of CRISPR”. In fact, though, the two discoveries aren’t remotely comparable.

On 23 September, Anthropic said it has turned its attention to biology and that its large language model, called Claude, has discovered “a novel enzyme system with CRISPR-like repeats”.

What Anthropic did was ask Claude to search through a large DNA database for “interesting new” proteins of a kind called a reverse transcriptase. These are enzymes that can make a DNA copy of an RNA sequence. Normally, DNA is transcribed into RNA, hence the reverse part.

Why is Anthropic interested in reverse transcriptases? Because if the aim is to genetically engineer anything from bacteria to people, being able to add specific DNA sequences to specific sites in a genome is really useful. There are already ways of doing this, but they all have limitations.

Claude agents found around 200,000 sequences in the database that probably code for reverse transcriptases. While analysing one family of reverse transcriptases, which are found in bacteria-infecting viruses, one agent detected a series of repeated sequences next to the sequence for the reverse transcriptase.

“I can see by eye a tandem repeat array… that’s a CRISPR-like… repeat array?!” the agent outputted.

What you need to know here is that CRISPR stands for clustered regularly interspaced short palindromic repeats. In other words, an area of DNA where there are lots of short repeated sequences separated by non-repeating DNA.

Such repeats were first spotted in E. coli in 1987 by researchers in Japan and later dubbed CRISPR by Francisco Mojica at the University of Alicante in Spain. In 2005, Mojica reported that the pieces of non-repeating DNA match bits of viral DNA.

He correctly deduced this must be some kind of bacterial immune system, but it took years of research by many different teams to uncover how it works, with a protein called CRISPR-Cas9 playing a key role. The real breakthrough came in 2012 when Jennifer Doudna at the University of California, Berkeley, and her colleagues worked out how to create pieces of RNA that can hook up with the CRISPR-Cas9 protein and guide it to destroy specific sequences. This was the start of CRISPR gene editing, which has revolutionised biology.

So it is really intriguing that Claude has found a similar set of repeats in bacteria-infecting viruses. It was already known that some viruses have CRISPR systems, but this is something different that is worth studying further.

In terms of significance, however, Claude’s work is comparable to the 1987 discovery in E. coli, not the 2012 breakthrough that won a Nobel prize. And with regard to AI contributions to biology, Claude’s work isn’t even close to being in the same league as DeepMind’s AlphaFold system for predicting protein structure, which is proving to be truly transformative.

Will Claude’s discovery lead to powerful new gene-editing tools? The context here is that we have discovered many CRISPR variants, along with a few other natural systems – most notably bridge recombinases – that could be useful for gene editing.

What’s more, proteins such as CRISPR-Cas9 have now been extensively altered and evolved in the lab to make them better for gene editing. For instance, base editors – which are already saving lives – are CRISPR proteins engineered to change a single letter at a time without cutting DNA strands, as the original Cas9 protein does.

Then there are prime editors, which consist of a reverse transcriptase fused with part of the Cas9 protein, that can add short sequences of DNA to genomes. In other words, we have already adapted reverse transcriptases for RNA-targeted gene editing.

It is impressive that Claude has identified what the Anthropic team is calling array-associated reverse transcriptases, or ART. And it is possible, if unlikely, that this viral system works in a unique way that allows us to do things that cannot be done with existing genetic tools. But there is a lot more work to be done to find this out, and at best the outcome will be one more gene-editing tool in a rapidly expanding set.