Can AI Invent Viruses to Fight Superbugs?
- Owen Coggins
- Sep 20
- 2 min read
Antibiotic resistance is one of the biggest threats to human health today. Many bacteria are becoming resistant to the drugs we normally use to kill them, which makes once-treatable infections dangerous or even deadly. One promising alternative is phage therapy, which uses bacteriophages — viruses that naturally infect bacteria — to infect and kill them. But there’s a catch: bacteria can evolve resistance to phages just like they do with antibiotics. This raises a big question: could artificial intelligence (AI) help us design brand-new phages, ones that don’t exist in nature, that can still infect bacteria and possibly outsmart resistance?
The Study Design
To explore this question, researchers turned to genome language models, a type of AI similar to ChatGPT but trained on DNA instead of English. They built two models, Evo 1 and Evo 2, trained on millions of DNA sequences, including more than two million phage genomes.
As a test case, they chose a small, well-studied phage called ΦX174, which infects E. coli. Here’s the basic process they followed:
They fine-tuned the AI on closely related phage genomes so it could “learn” the rules of phage biology.
The AI then generated hundreds of synthetic phage genomes that looked like ΦX174 in structure but with many DNA differences.
Computational filters removed sequences that looked too broken to work.
From the remaining candidates, the team synthesized around 285 phage genomes in the lab.
These designs were tested in bacterial cultures to see if they could infect and kill E. coli. They also tested whether these AI-made phages could defeat bacteria that had evolved resistance to ΦX174.
The Major Findings
The results were striking. Out of 285 designs, 16 of the AI-made phages came to life and successfully infected E. coli. Many of them had dozens to hundreds of mutations compared to any natural phages, proving that AI could generate true biological novelty.
What was even more impressive , some AI-designed phages performed better than the natural ΦX174 — killing bacteria more quickly or growing more efficiently. When bacteria became resistant to ΦX174, cocktails of the AI-made phages could still wipe them out, while ΦX174 alone failed completely.
One especially surprising result came when the AI swapped out a critical protein, known as the J protein, from a distant phage relative. Scientists had assumed this would break the virus, but the AI introduced other genetic changes that kept it working. This suggested the AI could discover creative biological solutions humans might never think of.
Why It Matters
This study shows that AI can do more than just analyze existing biology — it can invent new life forms that solve real medical problems. If developed further, AI-designed phages could provide:
New weapons against superbugs (antibiotic resistant bacteria) when antibiotics fail.
Faster innovation, since scientists wouldn’t need to wait for natural evolution to produce useful phages.
Personalized treatments, where doctors could order custom-designed phages for a patient’s exact infection.
For students and young scientists, this is a glimpse into the future of biotechnology. Just like ChatGPT can “write” essays, these genome models can “write” viruses. It’s an exciting example of how computer science, genetics, and engineering are coming together to tackle one of humanity’s toughest health challenges.