The use of artificial intelligence in fields like health care has been growing in recent years — and now scientists have used AI to build a virus that could kill dangerous bacteria like E. coli.

Scientists at Stanford University have created the first virus designed by AI. Researchers used a generative AI model called EVO 2 to create a specific kind of virus known as a “bacteriophage.”

The term “phage” is typically used as a suffix that means “to eat or to devour,” said Dr. Greg German, a physician, microbiologist and clinician investigator focusing on urinary tract infections and phage therapy at Unity Health Toronto.

“Bacterial phage eats bacteria, reprograms them to make more phage just like human viruses attack human cells to make more human viruses,” he said.

Why create a new virus?

Scientists hope bacteriophages can effectively function as new antibiotics, battling bacteria like E. coli that are harmful to humans.

“It’s essentially the enemy of my enemy,” German said.

“Bacteriophages and bacteria have been having a battle for over two and a half billion years. And now we’re trying to figure out the right ways to use them (for humans),” he added.

Similar research is underway in Canada, too.

Last month, Unity Health in Toronto was chosen to lead a $25 million international research trial on using viruses to treat drug-resistant infections.

Get weekly health news

German, who is leading the trial, says they will treat 212 women suffering from urinary tract infections due to E. coli over the next four years with phages.

Scientists have to isolate phages to design them to target a specific disease in a specific person, German said, adding that the phages that the women in the trial will be treated with will be “personalized to them.”

It’s like finding the right “key” to a “lock,” he said.

Research is also going on to determine how these phages can impact people who are suffering from other conditions such as prosthetic joint infections or cystic fibrosis, he added.

How did AI make a virus?

Stanford scientists built 300 novel phages and tested them for efficacy against E. coli. They found that 16 of them were good at killing E. coli.

The paper on the findings, published in the journal Science, said these genomes are “divergent from those observed in nature and with prespecified traits.”

This means that they are not naturally occurring and can be designed with a specific purpose in mind, in this case attacking E. coli.

For starters, the bacteriophage built by Stanford scientists is relatively small, German said.

While a typical bacteriophage has between 50,000 and 200,000 units of information in its genetic code, the phage built by EVO 2 has around 6,500, he added.

“They used language models and other predictive factors associated with modern AI to work through this and also not go too far off the beaten path,” German said.

The paper says the research “lays a foundation for AI-guided design of biological function at the whole-genome scale.”

Typically, phages are grown within the body of the bacteria, German said.

The Stanford process circumvented that and built the phage outside the “stinky container” of the bacteria.

“This changes the diagram and the flow. It takes $2 billion and 10 years of work to make a new antibiotic. We’re looking for a future in the next 10 years that after only eight hours, you’ve just made your therapy,” he said.

This will save scientists time in sorting through millions of parts of bacterial DNA and removing all the “bacterial junk” associated with preparing a phage.

“The theory is that you could take a swab of something or a urine sample, put it in a machine, and then — boop, boop boop — it decides what virus will be used to target this,” he said.

However, some experts are tempering the excitement with caution, citing safety concerns of letting artificial intelligence create the genome for a virus.

“Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions,” an accompanying article in Science said in response to the Stanford research.

“The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not,” it adds.