Scientists used AI models trained on millions of DNA sequences to generate complete bacteriophage genomes, synthesised nearly 300 designs and found 16 that produced functional viruses capable of infecting E. coli | Image used for representational purpose only
| Photo Credit: Getty Images
Scientists from Stanford University and the Arc Institute have used artificial intelligence to design complete genomes of bacteriophages viruses that infect bacteria, in a study that could advance research into alternatives to antibiotics for treating drug-resistant infections.
Published in Science, the study is among the first demonstrations that AI can design entire viral genomes rather than individual genes or proteins. The researchers generated thousands of potential bacteriophage genomes, chemically synthesised nearly 300 and tested them in the laboratory. Sixteen produced functional viruses capable of infecting Escherichia coli.
How did AI design them?
The researchers used genome language models called Evo 1 and Evo 2, which learn patterns in DNA sequences in a manner similar to how language models learn patterns in human language. Trained on millions of genomes, the models can capture patterns of genome organisation and constraints shaped by evolution.
The researchers asked the models to generate complete genomes for bacteriophages targeting a particular strain of E. coli, using the naturally occurring bacteriophage ΦX174 as a starting framework.
The viable designs were not simply copies of existing viruses. They contained new combinations of genes and regulatory elements and differed in genome length. One generated phage also contained a DNA-packaging protein that was evolutionarily distant from proteins normally associated with its capsid, the protein shell surrounding a virus.
The findings suggest that AI can generate previously unseen genetic combinations while retaining biological function.
Can AI overcome resistance?
Bacteria can develop resistance to bacteriophages, just as they develop resistance to antibiotics. The researchers therefore tested whether the AI-designed phages could infect E. coli strains resistant to ΦX174.
A cocktail of the newly designed phages rapidly overcame this resistance in laboratory experiments, whereas a comparable mixture of naturally sourced ΦX174-like phages did not.
The result suggests that AI could expand the range of phage designs available for testing and help identify combinations that remain effective as bacteria evolve.
However, the study does not demonstrate a treatment for human infections. The experiments were conducted under laboratory conditions using phages that infect bacteria, not humans. Their safety and effectiveness as therapies would require extensive further testing.
Challenges and concerns
Although the study involved bacteriophages that infect E. coli and cannot infect humans, experts in a related Science perspective warned that generative AI capable of composing viral genomes raises significant biosafety and biosecurity questions.
Similar technologies could potentially be applied to pathogens affecting humans, animals or plants, where biological behaviour may be difficult to predict or control.
Only 16 of the nearly 300 synthesised designs were viable. But the study shows that AI-generated sequences can be used to produce complete viral genomes with biological activity under laboratory conditions.
Any future medical applications will require extensive validation and safeguards to prevent misuse.
Published – August 08, 2026 11:13 am IST