Stanford AI Designs 16 Novel Bacteriophages to Kill E. Coli

Stanford AI writes complete genomes for bacteria-killing viruses

Stanford researchers have successfully used a generative AI model to design novel bacteriophages—viruses that kill bacteria—marking a significant advance toward AI-engineered antibiotics. The team synthesized and tested nearly 300 new phages based on AI-generated genomes, ultimately identifying 16 that proved exceptionally effective against E. coli.

Brian Hie, assistant professor of chemical engineering at Stanford, developed Evo 2, a generative AI model that writes complete DNA sequences from minimal starting information. Working with bioengineering graduate student Samuel King, Hie applied the model to bacteriophage ΦX174, a virus with a genome of fewer than 6,000 base pairs that naturally targets E. coli.

The AI generated thousands of candidate genomes in a single pass, writing complete sequences without human intervention. "We wanted the model to generate the entire genome end-to-end in a single left-to-right pass. We didn't add anything," Hie explained. Lab testing revealed that several of Evo 2's designs demonstrated higher fitness than the natural ΦX174 phage.

Why it matters

Bacterial resistance remains one of modern medicine's most pressing challenges. Traditional antibiotics lose effectiveness as bacteria evolve immunity over time. The Stanford team's approach addresses this vulnerability by creating genetically diverse phage cocktails. When bacteria develop resistance to one phage, others in the mixture remain effective. Lab tests confirmed that a cocktail of the 16 AI-designed phages rapidly overcame E. coli that had become immune to natural ΦX174. This strategy could extend to other dangerous pathogens including tuberculosis, MRSA, and hospital-acquired Pseudomonas aeruginosa infections.

From thousands of options to viable candidates

King, the paper's first author, developed a computational framework to evaluate the AI-generated genomes and narrow thousands of candidates to the most promising options. The framework assessed design criteria based on ΦX174 and related phages, then selected optimal candidates for chemical synthesis and laboratory testing.

The process involved substantial technical and financial challenges. Even with ΦX174's relatively short genome, interpreting a 5,400-character DNA sequence gene-by-gene requires sophisticated analysis. High DNA synthesis costs meant the team needed to focus resources on the most viable alternatives.

Open source availability and safety considerations

Hie has made Evo 2 freely available as open-source software, enabling researchers worldwide to design new genomes. This decision has sparked discussions about biosecurity risks. Hie argues that open availability accelerates beneficial research and that existing natural pathogens pose greater immediate risks than potential AI designs. He notes that AI tools allow researchers to build in safety checks—something impossible with naturally evolving pathogens—and provide powerful defenses against both natural pandemics and engineered biological threats.

Looking ahead, Hie is collaborating with researchers at Stanford and other institutions to extend Evo 2's capabilities to longer, more complex DNA sequences. Future applications could include engineering beneficial microbes that produce useful chemicals, medicines, and fuels. "The biggest open questions for me are how do we get greater genetic novelty and how do we get greater controllability of the outcomes?" Hie said.

These details were first reported by Stanford News. The research received funding from the Arc Institute, the National Science Foundation, the Knight-Hennessy Graduate Scholarship Fund, the Fannie and John Hertz Foundation, and the Stanford Institute for Human-Centered AI.