Stanford researchers have used the Evo 2 generative AI model to design new viruses that can kill E. coli bacteria. The team synthesized nearly 300 phages from DNA sequences produced by the AI, and laboratory testing narrowed that group down to 16 phages with particularly strong E. coli-killing activity.
According to Stanford News, the work centers on bacteriophage ΦX174, pronounced "FYE-ex-1-7-4." This is a type of virus that naturally infects bacteria, making it a potential tool for fighting bacterial infections.
How Evo 2 AI Designs New Phage Genomes
The Evo 2 model works by generating new DNA sequences from a small starting snippet of a phage genome. In this case, the researchers asked the AI to produce an entire ΦX174 genome in a single left-to-right pass — meaning the model wrote out the complete genetic code from start to finish.
Brian Hie, an assistant professor of chemical engineering and Dieter Schwarz Foundation Stanford Data Science Faculty Fellow, created Evo 2. Bioengineering graduate student Samuel King led the experimental work described in the paper, according to Science.org.
The research leverages genome language models — Evo 1 and Evo 2 — to generate complete phage genomes with realistic genetic architectures. This means the AI is not just randomly assembling DNA; it is learning the patterns of real genomes and producing sequences that look and behave like natural ones.
From AI Design to Lab-Tested E. coli Killers
After Evo 2 generated the DNA sequences, the researchers synthesized the actual phages in the laboratory. They created nearly 300 novel phages and tested them for effectiveness against E. coli, as reported by Stanford News.
The testing process narrowed the field significantly. Only 16 of the nearly 300 phages showed particularly strong E. coli-killing activity. This means the AI-generated designs were not all equally effective — but the successful ones demonstrate that the approach can work.
"In this case, we wanted the..." — Stanford News
This research represents a step toward AI-designed life. According to Nature, the work has been described as the world's first AI-designed viruses — a step towards AI-generated life. The phages created by the Stanford team are entirely new, not modified versions of existing viruses.
Why AI-Designed Phages Matter for E. coli Treatment
Phages are viruses that specifically infect bacteria. They are being studied as an alternative to antibiotics, especially for bacteria that have become resistant to traditional drugs. E. coli is a common cause of food poisoning, urinary tract infections, and other illnesses.
The ability to design phages with AI could speed up the process of finding new treatments. Instead of searching through natural environments for phages that happen to work against a specific bacterium, researchers can now generate many candidate phages quickly and test them in the lab.
According to Instagram, Stanford researchers used AI genome models to create 16 entirely new bacteriophages capable of infecting E. coli. The viruses were based on AI-generated designs, showing that the technology can produce functional biological agents.
Our Take: AI Biology Is Moving From Theory to Practice
This research is significant because it shows AI can do more than predict or analyze — it can create. The Evo 2 model designed complete viral genomes that actually worked when built in the lab. That is a meaningful step forward.
To put it plainly, this is not just a computer simulation. These are real viruses, synthesized from AI-written DNA, that killed E. coli in laboratory tests. The fact that 16 out of nearly 300 worked is a reasonable success rate for a first attempt at this kind of design.
There are important questions to consider. How will these AI-designed phages behave in living organisms, not just in lab dishes? Can the approach be scaled to target other harmful bacteria? And what safety measures are needed as AI becomes capable of designing biological agents?
For now, the Stanford work demonstrates a proof of concept. AI can design functional viruses. The next steps will determine whether this becomes a practical tool for medicine or remains a laboratory achievement.
