According to The New York Times, US startup Radical Numerics described in detail how it developed the virus in a technical report published on the company’s website on October 20. The company declined to provide information about the patient or the doctor who conducted the treatment, citing confidentiality concerns.
According to Futurism, although the report has not yet undergone peer review or been published in a scientific journal, company executives said the trial had been authorised by the US Food and Drug Administration (FDA). “We intend to publish our scientific findings in a specialist journal, while any future disclosure of clinical details will require consent from the patient or the relevant authorities,” said Eric Nguyen, CEO of Radical Numerics.
If successful, the experiment could open up a new avenue for applying AI in medicine, enabling personalised treatment plans designed to eliminate even the most stubborn antibiotic-resistant bacteria.
Although AI-designed viruses are a recent development, bacteriophage therapy has existed for more than a century. According to Chosun Daily, these viruses attach themselves to the surface of bacteria and inject their genetic material into the cells. Inside, the genetic code replicates continuously, causing the bacteria to swell, burst and die. Bacteriophages have attracted considerable attention as a potential alternative to antibiotics for fighting drug-resistant superbugs, thanks to their ability to target specific bacteria with exceptional precision without damaging other healthy cells in the body.
Today, as antibiotic resistance becomes an increasingly serious global problem, this treatment approach is experiencing a revival, although it is considerably more difficult to administer than conventional oral medication. Each type of bacteriophage can kill only certain bacterial strains. Therefore, once a suitable phage has been identified, doctors must test whether it is actually effective against the patient’s infection. Dr Marisa Azad, an infectious disease physician at The Ottawa Hospital in Canada who is collaborating with Radical Numerics, acknowledged that finding suitable bacteriophages is a time-consuming process.
Some researchers are trying to overcome this obstacle by searching for new bacteriophages in soil and water. However, the emergence of AI has accelerated the idea of designing customised bacteriophages capable of targeting the specific bacteria responsible for an individual patient’s infection.
In 2025, a research team at the Arc Institute in Palo Alto, California, developed an AI model called Evo 2 and trained it on millions of DNA sequences, including bacteriophage genes. The researchers asked Evo 2 to generate new genomes resembling those of a group of viruses that target bacteria. As a result, the AI generated 700,000 viral genomes. The researchers synthesised DNA molecules from 285 of these genomes and introduced them into bacteria. They found that 16 genomes produced viruses capable of surviving.
Three members of the Evo team, including Dr Nguyen, subsequently left to establish Radical Numerics. In June, the company launched a new AI model called Omnii, which operates according to several principles similar to those of Evo. Unlike models limited to DNA, Omnii can be trained on other aspects of biology, including protein structures and molecules that regulate gene activity. Radical Numerics is using Omnii to tackle a range of challenges, including designing cancer vaccines and identifying the risk of viruses causing pandemics. Among these initiatives is the first trial of AI-designed bacteriophage therapy.
Simulation of an AI-designed bacteriophage attacking antibiotic-resistant bacteria.
Image: Radical Numerics.
Researchers created several variants of the T7 bacteriophage. This virus, which only attacks E. coli bacteria, resembles a miniature lunar lander, with a spherical protein shell containing DNA and six legs extending outward. To attack a bacterium, T7 lands on its surface, uses its legs to attach firmly to the site, and then injects its DNA into the bacterial cell. Dr. Nguyen and colleagues said that Omnii generated new viral genes that altered the shape of T7’s legs. Although some of the newly designed legs differed significantly from their natural counterparts, Omnii’s bacteriophages were still able to infect bacteria.
After confirming that the approach worked, the research team collaborated with the doctors treating a patient with a persistent E. coli infection. The bacteria were resistant not only to antibiotics but also to all available bacteriophages. The team trained Omnii on information about the resistant strain and asked it to design a T7 variant with legs capable of attaching to the bacteria. The AI generated a genome for the new virus, and scientists created the corresponding DNA molecule and inserted it into bacteria. The infected bacteria then produced T7 bacteriophages capable of effectively attacking and killing other bacteria. After conducting safety tests, Radical Numerics presented the virus to the doctors, and the FDA authorized a trial.
Phage therapy can last several days or weeks, and doctors typically monitor patients for months to check for signs of recurrent infection. Therefore, it will take some time to determine whether the virus designed by Omnii works effectively.
If the trial is successful, Radical Numerics hopes to help accelerate phage therapy by significantly reducing the time needed to create new viruses tailored to the characteristics of each patient’s infection.
(Source: VnExpress)
