← Back to Home

AI-Designed Phages Used in a Patient, but Efficacy Against Drug-Resistant Infection Remains Undisclosed

An experimental preparation containing AI-designed phages has been administered to a patient through compassionate use. It eliminated bacterial isolates from the patient in the laboratory, but whether it can improve infection in humans remains an unanswered question in the public record.

By SURL BioNews

When bacteria develop resistance to antibiotics, finding viruses that can target them may offer another treatment approach. Radical Numerics announced on October 20 that a cocktail containing phages designed by its artificial intelligence model Omnii-1 had been administered to a patient with a drug-resistant bacterial infection. This brings AI-guided virus design into human use while leaving the most critical question unanswered: did the patient benefit?

According to the company, the intervention was authorized, physician-led, and conducted through compassionate use. Such arrangements allow seriously ill patients who lack suitable alternative therapies to receive investigational treatments under supervision. The company calls this the world's first use of AI-generated phages in a patient, but this “first” should still be treated as a company claim and cannot be equated with proven efficacy.

Phages are viruses that infect bacteria. They recognize specific structures on bacterial surfaces and replicate after entering the bacteria; lytic phages suitable for therapeutic use then cause the bacteria to rupture. This recognition ability offers the possibility of precise targeting but also creates a matching challenge: a phage that can kill one bacterial strain may not be able to target the bacteria in another patient. Researchers therefore hope to use AI to help identify and modify candidates, narrowing the scope of screening them one by one.

“AI-designed viruses” also does not necessarily mean writing an entire viral genome from scratch. The approach described in the company's technical documentation includes retaining the main genome of an existing phage while redesigning regions associated with specific functions. Omnii-1 learns from genomic and multiomics data, then combines phage genomes with publicly available data on phage–bacteria interactions to help generate variants and predict matches. Experimental confirmation of activity is still required.

The publicly available results directly related to this patient currently extend only to in vitro testing: the company said that the cocktail containing variants designed by Omnii-1 eliminated drug-resistant bacteria isolated from the patient's infection sample before it was administered to the patient. Reporting by The Register also clearly distinguishes these two stages. Citing privacy protection, the company has not disclosed the patient's clinical course or outcomes, so the available information cannot establish whether the infection improved or whether adverse reactions occurred.

The species of bacteria causing the patient's infection has likewise not been disclosed. Although the technical report discusses the model's predictive performance for Pseudomonas aeruginosa, this does not establish that it was the pathogen in this case; the model testing in the report and the patient's case need to be understood separately. Evidence of bacterial killing in vitro supports further research but is not yet sufficient to answer whether the preparation can act at the infection site after entering the human body.

The company also acknowledges that these therapies have not been approved for broad clinical use and that observations from a single patient cannot establish safety or efficacy. The significance of this event is that AI design has moved beyond the laboratory into supervised human use. For it to become a treatment option for more patients, assessable clinical outcomes and more systematic evidence of safety and efficacy are still needed.

References

  1. ZME Science
  2. Radical Numerics
  3. The Register