The Pulse
Radical Numerics Gives AI-Designed Phage Cocktail to a Patient
Radical Numerics says its AI-designed phage cocktail was given to a patient with an antibiotic-resistant infection under compassionate use. The company reports that the cocktail eliminated the patient’s bacterial isolate in laboratory tests
AI.info Team · Written with AI · fact-checked ·
Radical Numerics says it has administered an AI-designed phage cocktail to a patient with an antibiotic-resistant infection, but the company has not disclosed whether the treatment helped. The cocktail eliminated the patient’s bacterial isolate in laboratory testing before it was given under an authorized, physician-led compassionate-use pathway, according to the company’s October 20 announcement.
The announcement marks a first, the company says: clinical administration of phages designed with its Omnii-1 model. It is not evidence that the treatment worked in the patient. Radical Numerics says it is withholding the clinical course and outcome to protect the patient’s privacy, and cautions that observations from one case cannot establish safety or efficacy.
Omnii-1 designs met the patient’s isolate in the lab
Phages are viruses that infect bacteria, not human cells. Their potential as a treatment depends on matching a phage to the specific bacterial strain causing an infection. Radical Numerics describes current personalized phage therapy as a search process: researchers test existing phage collections against a patient’s bacterial sample, then consider other collections, environmental samples or engineering when they find no suitable match.
The company trained Omnii-1, a multimodal genome language model, to help identify and design phages for particular bacterial strains. Its announcement says the patient received a cocktail containing Omnii-1-designed variants after laboratory tests showed that the cocktail eliminated the patient’s antibiotic-resistant isolate in vitro. It does not name the bacterium, describe the infection, identify the administered phages or provide the treatment’s timing and dose.
Compassionate use permits investigational treatment outside a clinical trial for a seriously ill patient when satisfactory alternatives are unavailable, subject to regulatory oversight. The pathway makes the administration possible; it does not turn a single treatment into a controlled test of whether the model-designed therapy is safe or effective.
Separate experiments test the model’s design choices
Radical Numerics also reports laboratory evaluations of Omnii-1’s phage-design capabilities. In one set of experiments, the team used T7, a well-characterized phage that infects E. coli, to test designs involving its receptor-binding tail fiber protein. At 11 to 15 amino-acid changes from the original sequence, the company recovered 69.9% of its designed variants, compared with 4.8% of random controls.
Recovery is a proxy for viability, not proof that a phage kills the intended bacteria or works as a treatment. The company separately measured activity across three bacterial strains and found that the model’s rankings enriched for higher-activity candidates, though results varied by strain. Those experiments describe model development; the announcement does not say they were the tests used to select the patient’s cocktail.
One administration leaves the clinical question open
The distinction between a successful laboratory test and a patient outcome is central to the announcement. A cocktail that clears bacteria from a dish may still face biological and clinical hurdles in a person, and the company provides no outcome data with which to judge this case. Radical Numerics also states that its investigational therapies are not approved for broader clinical use.
“These investigational therapies have not been approved by regulatory bodies for broader clinical use, and observations from a single patient cannot establish general safety or efficacy.”
Aniketh Iyengar, Benjamin Siranosian, Arpan Sarkar, Avery Noonan, Michael Poli, Eric Nguyen and other Radical Numerics authors
The next practical test is not a new claim about AI design, but whether clinical teams can repeatedly match, validate and manufacture phages for individual infections under appropriate oversight. Radical Numerics says it is seeking clinical partners and aims to work with treatment teams to produce patient-specific doses when promising candidates have been validated. For this patient, the company has reported administration and an in-vitro result, but no clinical outcome.