Startup Says AI-Designed Viruses Treated a Patient With Drug-Resistant E. coli Infection

The effort builds on work at the Arc Institute in Palo Alto, Calif., where researchers in 2025 created an AI model called Evo 2 trained on millions of DNA sequences, including phage genes. They prompted Evo 2 to make new genomes similar to those in a group of phage species. The model generated 700,000 virus genomes. The team synthesized DNA molecules from 285 genomes and inserted them into bacteria; 16 produced viable viruses. Three members of the Evo team, including Dr. Eric Nguyen, left to found Radical Numerics.

In June, Radical Numerics unveiled an AI model called Omnii, based on some of the same principles as Evo. Omnii can be trained not just on DNA but also on other aspects of biology, such as the structures of proteins and molecules that keep genes turned on or off. The company is now using Omnii to tackle a range of challenges, including designing cancer vaccines and recognizing the risk that a virus may cause a pandemic, the report said. Among these efforts is the first trial of AI-designed phage therapy.

The researchers created variations of a phage called T7, which only infects E. coli. T7 has a ball-shaped protein shell that holds its DNA atop six leglike protrusions; it lands on a bacterium, anchors itself with its legs and injects its DNA into the cell. Omnii generated new viral genes altering the shapes of T7's legs, according to Nguyen and his colleagues. Some of the new legs were significantly different from natural ones, but Omnii's phages could still infect bacteria.

Once the researchers were confident their strategy had worked, they partnered with doctors treating a patient with a stubborn E. coli infection. The bacteria had resisted not just antibiotics but all the available phages. The researchers said they had trained Omnii with information about the resistant strain and told it to design T7 phages with legs that could grab onto the bacteria. After the AI generated genomes for new viruses, the scientists produced the corresponding DNA molecules and inserted them into bacteria. The bacteria became infected and produced T7 phages that could successfully infect and kill other microbes. After running tests to make sure the phages were safe, Radical Numerics presented them to the doctors. The FDA authorized a trial.

The research has not yet been peer-reviewed in a scientific journal, and there is no guarantee the AI-designed phages will work. If the approach proves successful, it could help speed phage therapy by cutting down the time it takes to create a new virus tailored to each patient's infection. "We think we can get it done in days," said Dr. Eric Nguyen, the chief executive officer of Radical Numerics.

Editor's Summary

Radical Numerics reported that AI-designed T7 phages were given to a patient with a drug-resistant E. coli infection under an FDA-authorized trial. The company declined to identify the patient or the doctors, and the work has not been peer-reviewed. If validated, the approach could shorten the time needed to tailor phage therapy to individual infections.