U.S. researchers have used generative artificial intelligence to design microscopic viruses capable of attacking and killing bacteria, offering a new approach to combat antibiotic-resistant infections. Stanford University and Arc Institute scientists successfully synthesized 16 bacteriophages using genome language models like Evo 2 to target Escherichia coli.
Medical science has long sought fresh arsenals against increasingly resilient microbes, and now generative artificial intelligence is stepping directly into the laboratory. Researchers from Stanford University and the Arc Institute turned their focus toward bacteriophage ΦX174, a virus known for targeting Escherichia coli, commonly known as E. coli.
Translating Evo 2 Generative AI From Computer Models to Real-World Labs
The effort centers on Evo 2, a generative artificial intelligence model created by Stanford assistant professor of chemical engineering and Dieter Schwarz Foundation Stanford Data Science Faculty Fellow Brian Hie alongside bioengineering graduate student Samuel King. Until recently, the technology remained confined to computational environments. The team transitioned the model to physical experiments by synthesizing nearly 300 novel phages based entirely on genomes written by the AI.
Researchers evaluated the generated options and narrowed the candidate list down to a select group of 16 exceptionally effective E. coli-killing phages. The process relied on letting the model generate entire genetic sequences without human intervention during the writing phase.
“In this case, we wanted the model to generate the entire genome end-to-end in a single left-to-right pass. We didn’t add anything.”
Brian Hie, chemical engineer at Stanford University in California, via Mirage News
Laboratory testing confirmed that a few of Evo’s suggestions had higher fitness than the native ΦX174 organism.
Overcoming Bacterial Resistance Through Genetically Distinct Phage Cocktails
The choice of ΦX174 stems from its manageable scale. Featuring a genome under 6,000 base pairs long, it offers a simplified blueprint compared to the 3 billion base pairs found in the human genome, providing an ideal test case for artificial intelligence design capabilities.
Modern antibiotics routinely lose effectiveness over time because heavy usage drives microbes to evolve immunity. Relying on a single medication creates a distinct vulnerability in treatment protocols.
“If the bacteria gain resistance to a single phage, it’s game over for the medication. But if you have multiple genetically distinct phages in a mixture, it would be harder for the bacteria to develop resistance to the entire cocktail.”
Brian Hie, chemical engineer at Stanford University in California, via Mirage News
By combining multiple genetically diverse bacteriophages, scientists hope to develop resistance-resistant treatments. Similar strategies could eventually target other dangerous pathogens, including tuberculosis, methicillin-resistant Staphylococcus aureus, and Pseudomonas aeruginosa, which frequently causes hospital-acquired infections.
Computational Frameworks and Open-Source Tools
Constructing entire genomes nucleotide by nucleotide and introducing them into bacteria presents substantial technical hurdles and financial costs. Analyzing a 5,400-character string gene by gene is extraordinarily difficult for human researchers.
To solve this, Samuel King developed a computational framework enabling the team to evaluate generated options, establish specific design criteria, and select optimal candidates before chemical synthesis. This approach helped the team bypass expensive trial-and-error runs by focusing exclusively on viable alternatives.
To encourage broader scientific momentum, Hie offers Evo 2 open source and free of charge, allowing external researchers to download the model and design new genomes directly. While free availability sparks necessary discussions regarding biological safety and security, developers maintain that open access remains vital for accelerating real-world medical discoveries.
Related reading
