A team of researchers from Stanford University and the Arc Institute in Palo Alto, California say they’ve created viruses with AI-designed DNA that can target and kill specific bacteria.
And these aren’t just simulated possibilities — they’re real and already slaying germs in the lab.
The work, published in a new study awaiting peer-review, is a compelling testament to the usefulness of large language models in bioengineering applications, the authors say.
“This is the first time AI systems are able to write coherent genome-scale sequences,” senior author Brian Hie, a Stanford computational biologist, told Nature. “The next step is AI-generated life.”
Coauthor Samuel King, however, cautioned that a “lot of experimental advances need to occur in order to design an entire living organism.”
Viruses aren’t quite considered to be alive. Think of them like pesky little genomic robots that hijack our biology to replicate, since they don’t generate their own energy and can’t reproduce on their own. They aren’t made of cells, and are driven by a ruthless set of programmed instructions to multiply at all costs. Since their genomes are pretty simple, they’re easier to tinker with and less ambitious for a human or machine to recreate. Remember: a genome is all the DNA in an organism, not just a few strands.
In the study, the researchers used an AI model called Evo to dream up the virus genomes. Unlike a general purpose large language model, Evo is specifically trained on millions of bacteriophage genomes.
A bacteriophage is a virus that infects bacteria. To use as a jumping off point, the researchers chose a phage called phiX174 (or ΦX174), which infects strains of the wide family of bacteria known as E. coli. As the first DNA-based genome to be sequenced, phiX174 is an extensively studied and well understood virus, and only has around 5,400 base pairs and 11 genes, according to Nature.
After probing the AI model, the team came up with 302 virus designs. The best way to test them, the researchers figured, was to print, or chemically assemble, all of them and unleash them on real strains of E. Coli.
As it turned out, some of them worked. Once inserted into the poor waiting germs, 16 of the AI-designed viruses successfully infected their hosts by inserting their DNA, hijacking the bacteria to start cranking out copies of themselves, and then burst through the cell’s body, killing it.
Sixteen out of 302 isn’t the highest batting average, but it’s a remarkable achievement nonetheless. Overall, the researchers found that their bot-conceived creations could kill three different E. Coli strains, outperforming natural phiX174.
“In many cases,” wrote Niko McCarty, a former bioengineer for Caltech and Imperial College London, for Asimov Press, “they were more infectious than wildtype phiX174 despite carrying major genome alterations that a human would be unlikely to rationally design.”
As promising as the results are, they also raise huge ethical concerns. If an AI model can come up with functioning phages, it could also be potentially abused to create bioweapons, experts warned, or even unintentionally create an out-of-control virus.
“One area where I urge extreme caution is any viral enhancement research, especially when it’s random so you don’t know what you are getting,” Craig Venter, founder of the J. Craig Venter Institute, who is known for his pioneering work in creating organisms with synthetic DNA, told MIT Technology Review. “If someone did this with smallpox or anthrax, I would have grave concerns.”
And according to Venter, the AI usage isn’t all that radical — it’s “just a faster version of trial-and-error experiments.”
But until now, AI models have only demonstrated that they could generate some DNA sequences like in proteins. And some of these new AI-written genomes are “so distinct from any known bacteriophage genome that they would technically be classified as their own species,” according to McCarty.
In all, this is the first time the tech has produced entire genomes that actually work in the real world. And that’s pretty significant, good or bad.
“It was quite a surprising result that was really exciting for us because it shows that this method might potentially be very useful for therapeutics,” King told Nature.
More on synthetic biology: Scientist Warns That New Synthetic Lifeform Could Spell Doom for Humankind