In a groundbreaking development, scientists have used artificial intelligence (AI) to design the first-ever viruses capable of infecting and killing strains of Escherichia coli (E. coli). The achievement, reported by Nature, marks a significant step toward AI-driven biotechnology and potential new therapies to combat antibiotic-resistant bacteria.
“This is the first time AI systems are able to write coherent genome-scale sequences,” says Brian Hie, a computational biologist at Stanford University in California. “The next step is AI-generated life,” he adds, while cautioning that much more experimental progress is needed before a living organism could be fully designed by machines.
The study, led by Hie, Samuel King, and colleagues, was posted on the preprint server bioRxiv on 17 September and has yet to undergo peer review. Still, the findings highlight how AI could revolutionize phage therapy, a strategy that uses viruses to fight bacterial infections. “Hopefully, a strategy like this can complement existing phage-therapy strategies and someday augment therapeutics to target pathogens of concern,” says Hie.
How the AI Built Genomes
Previous AI tools have generated DNA sequences, proteins, and molecular complexes, but designing a complete genome has remained out of reach due to its complexity. To tackle this, the researchers used two AI models — Evo 1 and Evo 2 — trained on more than 2 million phage genomes. They then guided the models to create viral genomes based on ΦX174, a simple single-stranded DNA virus with 11 genes.
From thousands of AI-generated genomes, the team narrowed the list to 302 viable candidates. Laboratory experiments showed that 16 of these AI-designed bacteriophages successfully infected and killed E. coli strains, including types resistant to antibiotics. Notably, some of the engineered viruses outperformed the naturally occurring ΦX174 virus by targeting strains it could not affect.
“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,” says King.
Promise and Concerns
Experts view the study as both a scientific milestone and a source of ethical debate. “This study provides a compelling case study of what is possible today and sets the stage for more-ambitious applications in the future,” says Peter Koo, a computational biologist at Cold Spring Harbor Laboratory in New York.
Still, concerns remain about the dual-use dilemma — the possibility that the same AI tools could be misused to create harmful viruses. Kerstin Göpfrich, a synthetic biologist at Heidelberg University in Germany, notes that this issue is not unique to AI. “In research in general you always have a dual-use dilemma. There’s nothing specific about AI, and you can always use progress for the better or for the worse,” she says.
The researchers emphasized that they excluded human-affecting viruses from their training data and used only non-pathogenic strains with a strong safety record in molecular biology. They hope their approach will pave the way for safe, AI-designed viruses to address public health threats, particularly the growing crisis of antibiotic resistance.
“I think this will definitely be a growing field and I’m super excited about it,” says Göpfrich.
