Scientists use AI to design new viruses that kill E. coli in the lab

In short: Researchers trained AI on viral DNA and used it to design full virus genomes, and 16 of the lab-built designs became working viruses that kill bacteria.

What happened

A research team led by Brian Hie at Stanford University and the Arc Institute trained AI models called Evo 1 and Evo 2 on about 2 million bacteriophage genomes. Bacteriophages, or phages, are viruses that infect bacteria, not people.

Instead of learning from words, the AI learned from DNA sequences. You can think of it like teaching a writing tool by showing it millions of recipes, except the “recipes” are genetic instructions and the “meal” is a virus.

The scientists then asked the AI to generate complete genomes related to a well-studied phage called ΦX174, which infects E. coli. They produced about 302 candidate genomes, chemically synthesized the DNA, and assembled the viruses in lab conditions.

Out of those 302 AI-designed genomes, 16 became viable phages. That means they could infect E. coli, copy themselves, and kill the bacteria. Some appeared to kill faster or work against E. coli strains that resisted the natural version used as a reference.

Why it matters

This work suggests AI can help design whole viruses that function, at least in controlled lab settings and for viruses that target bacteria. Supporters say this could one day help create phages to fight antibiotic-resistant infections, as an alternative when antibiotics fail.

At the same time, experts note a safety concern: if AI can propose many workable viral genomes, it could lower the barrier for designing harmful biology in the future. Current reporting also stresses that this is not the same as designing human-infectious viruses with pandemic potential.

Source: NYTimes