← Back to Home

AI-Designed Phages Have Been Used in a Patient, but No Public Evidence Yet Shows Efficacy Against Drug-Resistant Infection

Radical Numerics says a cocktail containing AI-designed phages has been administered through compassionate use, but the patient’s clinical course has not been disclosed, and laboratory bacterial-killing results cannot yet establish whether the treatment worked. A subsequent statement also acknowledged the collaborating team’s leading role.

By SURL BioNews

When neither antibiotics nor existing phages offer a way forward, can new phages be designed for a patient with an infection? On October 20, startup Radical Numerics announced that a cocktail containing phages designed by its artificial intelligence model had been used in a patient with a drug-resistant bacterial infection. However, the announcement currently supports only the milestone of administration to a human; no public clinical results are available to assess whether the patient’s infection was brought under control.

Phages are viruses that infect bacteria and are highly specific in their recognition of hosts. This gives them the potential to target pathogenic bacteria precisely, but also makes suitable candidates difficult to find. According to a New York Times report published on Carl Zimmer’s website, the E. coli infecting this patient was reportedly resistant to antibiotics and could not be targeted by the phages available at the time.

Radical Numerics uses its Omnii-1 model to propose phage genetic variants, which are then tested experimentally. Using the T7 phage as an example, the company modified sequences associated with tail fiber proteins, which help the virus attach to bacteria, in an attempt to change the bacterial strains it can recognize. AI design here does not mean creating an entire virus from scratch, but modifying specific functional regions of an existing phage. The model combines genomic data with data on phage–host interactions to help select candidates worth testing.

The company said a cocktail containing variants designed by Omnii-1 cleared bacterial isolates obtained from the patient in in vitro testing and was subsequently administered through a physician-led compassionate use pathway. The New York Times reported that company personnel said they had obtained authorization from the U.S. Food and Drug Administration; authorization for an individual case does not mean the therapy has been approved for general clinical use. Citing privacy protection, the company did not disclose the patient’s clinical course or outcomes.

Laboratory bacterial-killing performance and whether the patient benefited therefore remain two separate questions. According to independent reporting on October 20, the company’s technical report had not yet been peer-reviewed or published in a scientific journal at that time. Microbiologists interviewed cautioned that in vitro activity cannot directly predict whether an infection can be controlled in the human body. Months of follow-up after treatment may also be needed to determine whether the infection recurs.

The attribution of collaborative contributions in the announcement was later supplemented. In a statement issued on November 9, CEO Eric Nguyen acknowledged that Tabula Bioscience had played a leading role in the overall treatment effort, providing significant scientific, technical, and AI contributions. He also clarified that the phages designed using Radical Numerics’ model were one component of the treatment cocktail. The two parties had disagreed over the timing and manner of the announcement, and the partner’s contributions should have been explained more clearly.

This case brings AI-assisted phage design to the threshold of human application while leaving clear limits on the evidence: the cocktail’s in vitro results alone cannot establish the clinical contribution of its AI-designed components, and a single patient is insufficient to establish general safety and efficacy. Whether this technology can shorten the wait for a matched treatment and improve outcomes for hard-to-treat infections still awaits clinical data open to scrutiny and more comprehensive research.

References

  1. Yahoo
  2. Radical Numerics
  3. The New York Times, reprinted on Carl Zimmer's website
  4. Radical Numerics