← Back to Home

AI-Designed Phages Enter Human Use: US Startup Says They Have Been Administered to a Patient With a Drug-Resistant Infection

Radical Numerics announced that a cocktail containing AI-designed phages has been administered through compassionate use. Laboratory results showed it could kill the patient’s bacterial strain, but whether the patient improved as a result remains publicly unanswered.

By SURL BioNews

When antibiotics struggle to control an infection, can viruses that specifically target the disease-causing bacteria be found for the patient? US startup Radical Numerics announced on October 20 that a cocktail containing bacteriophage variants designed by its AI model Omnii-1 had been administered to a patient with a drug-resistant bacterial infection. This brings efforts to design biological entities with AI into human use, while leaving the most important question unanswered: what effect did the treatment actually have?

The company says this is the world’s first use of AI-designed phages in a human, with administration taking place through an authorized, physician-led compassionate use pathway. Such arrangements give seriously ill patients who lack suitable alternative therapies an opportunity to receive investigational treatment outside clinical trials; they do not mean the preparation has been approved for general clinical use. The claim of a “first” should also remain attributed to the company at this stage.

The public evidence remains at two distinct levels: the company says the cocktail first eliminated the drug-resistant bacterial strain obtained from the patient in laboratory experiments and was then administered to the patient. However, bacterial killing in a culture environment cannot directly establish whether an infection has been controlled in the human body. Citing privacy protection, the company has not disclosed the patient’s clinical course or treatment outcomes, and has explicitly stated that observations from a single patient are insufficient to establish general safety or efficacy.

Bacteriophages are viruses that infect bacteria. Their ability to recognize specific bacterial strains is both therapeutically appealing and a challenge when matching them to a target. Omnii-1 aims to help researchers select and design suitable variants. The methods demonstrated by the company include modifying genomic regions involved in recognizing bacterial surfaces, then synthesizing candidates and experimentally testing their activity, rather than creating an entire viral genome from scratch each time.

The model was trained using publicly available phage–host interaction data, combined with activity measurements from design experiments. However, these platform validation results cannot fill the gaps in patient data. The Register noted on October 22 that the species of bacteria infecting the patient had not been disclosed, and that the separate Pseudomonas aeruginosa prediction experiments listed in the announcement did not establish that this was the bacterium infecting the patient. The outlet said the company had not responded to its questions.

Background

Previous research has used genomic models to design phages capable of infecting Escherichia coli and reproducing, demonstrating that computer-generated sequences can have biological functions. This announcement takes the question a step further: from viruses that function in the laboratory to an investigational preparation tailored to a patient’s bacterial strain and actually administered.

What is needed next is clinical evidence that can be evaluated. Whether the infection improved, how long any effect lasted, and whether adverse reactions occurred cannot be determined from the current announcement. This administration offers an indication of progress toward clinical application; turning individual attempts into a treatment that can be used repeatedly will still require scrutiny based on more complete data on safety, efficacy, and manufacturing quality.

References

  1. Hindustan Times
  2. Radical Numerics
  3. The Register