← Back to Home

AI-designed phages used in a patient with a drug-resistant infection; clinical outcomes remain undisclosed

From finding bacteria’s natural enemies to using AI to redesign how phages recognize bacteria, personalized anti-infective therapy has entered human use. But between killing bacteria in the laboratory and a patient’s recovery lies clinical evidence that has yet to be released.

By SURL BioNews

In tackling drug-resistant infections, the challenge is sometimes not simply a lack of drugs, but the inability to find a treatment that works against the particular bacterial strain at hand. Biological AI company Radical Numerics announced that phage variants designed by its research team using the Omnii-1 model have been administered to a patient as part of a mixture. This takes AI-assisted biological design beyond the laboratory, while leaving the most important question unanswered: how did the patient fare after treatment?

The company announced the news on October 20, calling it the world’s first use of AI-designed phages in a human. According to its announcement, treatment took place through an authorized, physician-led compassionate-use pathway. Such arrangements give seriously ill patients without suitable alternative therapies an opportunity to receive investigational treatment outside formal clinical trials. A company press release distributed through Business Wire also stated that researchers completed the phage design for the patient’s persistent drug-resistant infection within weeks.

Phages are viruses that infect bacteria. Lytic phages suitable for therapeutic use can replicate inside bacteria, eventually causing them to burst and die. Their specificity in recognizing bacterial strains is both an attraction and an obstacle: a phage capable of infecting one bacterial strain may not work against another strain of the same species. Traditional approaches often require screening existing collections or environmental samples one by one to find suitable candidates.

Omnii-1 seeks to shorten this search. According to the company’s technical description, the model first learns from genomic and multi-omics data, then is trained on phage genomes and experimentally measured phage–bacterium pairing data. AI design also does not have to involve writing an entire virus from scratch: in a publicly released demonstration using the T7 phage, the team redesigned regions of the tail fiber protein involved in recognizing bacteria, then used experiments to check whether the variants could form active phages.

For this patient, the company said that the mixture containing AI-designed variants eliminated drug-resistant bacteria isolated from the patient in in vitro tests before it was delivered for clinical use. This result supports the mixture’s laboratory activity against that strain, but it cannot directly establish that the infection was brought under control in the patient, nor can it independently distinguish the contribution of each component in the mixture.

Major gaps remain in the public information. Citing privacy protection, Radical Numerics has not disclosed the patient’s clinical course or treatment outcomes. The Register also noted that the technical report did not identify the bacteria causing the patient’s infection; the Pseudomonas aeruginosa experiments in the report cannot be used to infer the patient’s diagnosis. Although the press release quoted comments on the approach from Ottawa Hospital physician-scientist Marisa Azad, it did not specify where the treatment actually took place.

Both phage therapy and engineered phages have precedents. What is new here is that candidates designed with AI involvement have entered human use. However, authorization for compassionate use does not mean the therapy has been approved for widespread use. The company also explicitly stated that a single patient cannot establish general conclusions about safety and efficacy. Whether this approach can truly shorten patients’ wait for treatment and deliver reproducible clinical benefits remains a question for publicly released treatment data and further research.

References

  1. TechRadar
  2. Radical Numerics
  3. Radical Numerics via Business Wire, republished by FinancialContent
  4. The Register