Anthropic’s AI biolab finds ‘CRISPR-like’ DNA in viruses. What’s next?

Sure, AI can parse petabytes of data. But how well can it pipette?

Regardless of the answer, artificial intelligence titan Anthropic has launched a biology ‘wet lab’ where human scientists and AI agents will work together to design and conduct experiments. This week, the company announced one of the team’s first finds: a peculiar pattern of DNA in the genomes of several giant viruses.

Similar patterns are found in DNA encoding the CRISPR immune systems of some microbes. Scientists have developed those microbial systems into powerful genome-editing tools. But researchers at Anthropic, which is headquartered in San Francisco, California, have not yet determined what the newfound viral sequences do, and thus whether their function is similar to that of the parallel sequences in microbial CRISPR systems.

In search of ‘weird things’

Even so, the finding, posted online on the alphaXiv platform, offers a first glimpse into Anthropic’s new life sciences research group and its associated lab, as well as how its AI tools could be used to mine petabytes of genomic data in search of novel molecular tools. The company announced the new lab and released the preprint on 23 September. The preprint has not yet been peer reviewed.

CRISPR gets a power boost from AI-designed ‘molecular scissors’

Many of the most impactful findings in biology have come from “finding weird things” in microbes, says Eric Kauderer-Abrams, head of life sciences at Anthropic. “Our goal was to see if we could systematize and scale up that kind of research.”

This initial experiment enlisted roughly 950 AI agents, which are autonomous AI systems that often rely on large-language models (LLMs). Anthropic’s agents spent more than 21 hours exploring billions of proteins in a self-directed fashion, discussing preliminary results and deciding on next steps amongst themselves.

In mathematics, a spate of AI-generated solutions to long-standing math problems has raised questions about the future role of human experts in the field. But the human contribution to biology remains clear for the foreseeable future, says Kauderer-Abrams. Life sciences research “requires actually running experiments in the physical world,” he says. “Biology is hard.”

Repeat finding

To start, Anthropic’s researchers told its AI agents to survey a database of DNA sequences encoding billions of proteins. The agents were directed to search for proteins that might work together with enzymes called reverse transcriptases. While examining the DNA surrounding a reverse transcriptase gene in a giant virus, the agents spotted the same short sequence of DNA letters over and over again. The agents decided the find was worth a closer look and ultimately found similar patterns in the genomes of other viruses.

The arrays of repeated DNA they found are somewhat like those seen in CRISPR systems. A bacterial CRISPR system, for example, might contain multiple repeated chunks of DNA. These are separated by segments of DNA copied from viruses or other invaders. When an invader arrives, RNA made from those DNA segments then helps to guide a partner enzyme to cut the interloper’s DNA, slicing the intruder’s genome into bits.

Can Anthropic’s invisible watermarks curb ‘AI slop’? Researchers remain sceptical

However, there is scant evidence that the repeats found by Anthropic’s team of AI agents carry out similar functions, and there is no known DNA-slicing enzyme partnered with them. “This is a promising lead,” says Kauderer-Abrams. “What’s hard is to then go and completely characterize it, understand its function, and develop it into an interesting tool.”