Artificial intelligence can dash off more than emails and essays. It has now written entire genetic instruction manuals, or genomes. 

Two AI models designed complete sets of DNA for 16 viruses that can attack Escherichia coli bacteria in lab dishes. Such viruses that infect and replicate inside bacteria are known as bacteriophages (Bak-TEER-ee-oh-FAHZ-es). Just call them phages.

A mix of these AI-designed phages stopped the growth of E. coli strains that ordinarily can resist death by virus. This hints that AI-made phages might offer new ways to fight tough-to-treat infections.

It’s the first time that AI has created an entire genome, says Brian Hie. And even though scientists debate whether viruses are alive or not, the work is a step toward using AI to design living things.

Hie is a computational biologist at Stanford University in California. He’s part of a team that shared its new findings September 17 on bioRxiv.org. (Studies posted to that site have not yet been vetted by other scientists.)

Writing in the language of DNA

In the past, AI models have been used to design individual genes and proteins. Creating an entire genome from scratch, however, adds a new layer of complexity. It requires many genes and proteins to team up.

To try to create genomes for bacteria-killing viruses, Hie’s team turned to two of its own AI models. These models had trained on billions of pairs of DNA’s basic units — abbreviated A, C, G and T — from existing phage genomes. (This is similar to the way ChatGPT has trained on the billions of letters in books and internet posts.)

The researchers turned to a known bacteria-killing phage — ΦX174 — as a guide to help their AI design a similar genome. Scientists first read out the complete set of genetic instructions for ΦX174 in 1977. Its DNA is now very well known and studied. This meant the scientists could easily compare how phage genomes designed by AI differed from that of ΦX174, Hie explains.

One reason his team decided to test their genome-designing AI on phages: They don’t infect people. That means it was safe to work with them in the lab. The researchers also did not train their models on viruses known to cause disease. In this way, they made sure their AI wouldn’t design viruses that might harm people.

Bacterial assassins

The two AI models came up with roughly 300 potential phage genomes. Of those, 16 produced viruses that could infect E. coli. Some of them even killed E. coli faster than ΦX174 did.

What’s more, ΦX174 failed to kill three strains of E. coli on its own. But mixtures of AI-generated phages quickly evolved to take out even those bacteria. 

The new findings suggest AI could help develop new viruses for use in medicine. Indeed, such phages might offer a way to treat bacterial infections that no longer respond to antibiotics.

“The need to find a phage that targets [such a ‘superbug’] strain would be very urgent,” says Kimberly Davis. “AI could be a powerful way of rapidly generating a phage match to treat patients.” Davis is a microbiologist at the Johns Hopkins Bloomberg School of Public Health in Baltimore, Md. She did not take part in the AI genome work.

Looking ahead

However, Davis adds, “Use of AI-generated phages would need to be tightly controlled.” Extensive tests would need to make sure such phages don’t harm good microbes in the body, ones that keep us healthy.

Ideally, AI-made phages would not just kill one harmful type of bacteria, Hie says. They might also evolve to keep up with virus-resistant bacteria. Using AI to design entire organisms also could make microbes that speed up antibiotic production, Hie notes. Or they might allow the design of microbes to break down plastics. 

AI may even help researchers make sense of genomes — and treat diseases — that are more complex than viral ones, Hie says. Our genome is more than half a million times the size of ΦX174’s genome, he notes. “So there’s a lot of work to go.”