AI Creates Novel Bacteriophage Genomes to Fight Drug Resistance

AI-Designed Viruses Show Promise Against Antibiotic Resistance

Researchers from Stanford University and the Arc Institute have successfully used artificial intelligence to create novel viral genomes capable of targeting specific bacterial hosts, according to a study published Thursday in the journal Science. The breakthrough demonstrates AI's potential to accelerate therapeutic development while simultaneously highlighting urgent governance gaps in biotechnology.

The team trained an AI tool called Evo on genetic sequences from millions of sources spanning all domains of life, similar to how language models like ChatGPT learn from text. The system then generated thousands of genome combinations designed to infect E. coli bacteria cells. Of approximately 300 genomes physically constructed and tested in the laboratory, 16 proved viable as functioning viruses.

These new viruses are bacteriophages—organisms that infect bacteria but cannot affect humans. Critically, tests revealed that a mixture of the AI-generated phages could overcome antibacterial resistance in certain E. coli strains where naturally sourced phages failed. The researchers noted that one virus exhibited genetic features "evolutionarily distant" from natural examples, suggesting the AI identified viable configurations that might have required millions of years of natural evolution to emerge.

Why It Matters

As antibiotic resistance becomes a growing global health threat, phage therapy represents a promising alternative treatment avenue. This research demonstrates that AI can potentially design adaptive therapies faster than traditional methods, directly addressing the challenge of rapidly evolving pathogens. However, the same capability that makes this technology medically valuable also creates biosecurity risks if applied to human-infecting pathogens.

Biosecurity Concerns Accompany Scientific Progress

Experts from the Johns Hopkins Center for Health Security, writing in a corresponding Science article, emphasized that while the capability to compose viral genomes using generative AI now exists, appropriate governance frameworks do not. They specifically warned against applying these techniques to eukaryote-infecting pathogens, which cause infections like malaria in humans, noting such work "might encode new pathogens that can infect humans, animals, or plants in ways that cannot be contained by existing countermeasures."

The Stanford researchers addressed biosecurity considerations in their work more thoroughly than most developers of biological AI models, according to the Johns Hopkins response. Still, the fundamental tension remains unresolved.

Jordi García Ojalvo, a systems biology professor at Pompeu Fabra University of Barcelona, offered a more measured assessment of immediate risk. He noted the low efficiency rate—just 16 viable viruses from hundreds of thousands generated—and the requirement for individual laboratory testing means these models cannot automatically produce dangerous pathogens "out-of-the-box."

The research focused specifically on bacteriophages targeting E. coli, and it remains unclear how effectively the approach would translate to other viral types. The authors characterized their work as laying out "a path for generating adaptive and resilient phage therapies against rapidly evolving pathogens."

These details were first reported by CNN.