AI-Designed Phage Treatment of Resistant Bacterial Infection

A biopharm start-up, Radical Numerics, describes using AI to design a phage for personalized treatment of a patient with resistant bacterial infection.¹ The New York Times reported that the pathogen was E coli which had proved resistant not only to antibiotics but to available phage therapies.²

The report issued by the company, albeit not peer-reviewed, could indicate that the arduous task of hunting for phage virus "needles in a haystack" (or in wetlands, soil and other environments),³ could be obviated by designing them.

The report lead author Aniketh Lyengar and colleagues describe challenges in developing successful phage therapy which go beyond isolating a phage which recognizes the molecular configuration of a receptor site on the surface of a particular strain of infecting bacteria.

"Recognition is only the first challenge," they indicate. "After attachment a phage must deliver its genome, overcome bacterial defenses, and replicate inside the cell to produce new phages. Failure at any of these steps can prevent a phage from establishing a productive infection, even after successfully recognizing a bacterium."

These interactions, Lynegar and colleagues remark, "make phage engineering a challenging biological design problem".

"Changes to a phage's genome can alter which bacteria it recognizes but may also disrupt functions required for assembly or replication," they explain."Successful engineering requires improving activity against the target bacteria while preserving the machinery necessary for infection."

The AI tool, Omnii-1, that they task with identifying phage genome targets to edit, was trained on datasets of experimentally measured phage-host interactions from the company's libraries and others that are publicly available.⁴ The tool itself was tested against two other specialized AI machine learning systems for phage-host prediction, with the AI-designed phages pitted against an E coli clade comprising 37 strains that had not been encountered in previous training.

"Omnii-1 maintained higher rates of identifying active phages than both specialized workflows across the top five recommendations, reaching 91.4% of eligible strains with just two phages," Lynegar and colleagues report.

From among the functional phage candidates developed with Omnii-1, the investigators identified those with higher activity against specific bacterial strains for further experimental testing. Candidates were ranked by predicted fitness and their measured activity compared across three bacterial strains.

"Beyond individual phage design campaigns, developing personalized therapies at scale requires understanding phage-host interactions across diverse bacterial strains," Lynegar and colleagues note.

The particular phage administered under Compassionate Use designation from the FDA for the patient with E coli is described by Carl Zimmer in the New York Times² as looking like a "microscopic Apollo lunar lander: a ball-shaped protein shell that holds its DNA atop six leg-like protrusions." The number and shape of those legs, altered through newly generated viral genes following Omnii-1 guidance, ensured the phage attachment to the pathogen.

Zimmer reports that the company intends to present the findings for peer review, with the caveat that disclosure of clinical details will be subject to appropriate consent. Lynegar and colleagues elaborate in the technical report:

"To protect the patient's privacy, we are not disclosing details of their clinical course or outcomes, as even limited clinical information could make the patient identifiable," they state."These investigational therapies have not been approved by regulatory bodies for broader clinical use, and observations from a single patient cannot establish safety or efficacy."

References

1. Lyengar A, Siranosian B, Sarkar A, et al. Radical Numerics creates first AI-generated phages to reach human patient with Omnii-1. Radical Numerics Report. 2026, October 20. https://www.radicalnumerics.ai/blog/omnii-phage.

2. Zimmer C. Start-up claims to use A.I.-created viruses to treat a patient. New York Times. 2026, October 20. https://www.nytimes.com/2026/10/20/science/ai-virus-bacteria-radical-numerics.html.

3. Pope WH, Jacobs-Sera D, Russell DA. Expanding the diversity of mycobacteriophages: Insights into genome architecture and evolution. PLoS One. 2011, Jan 27; 6(1):e16329. doi:10.1371/journal.pone.0016329.

4. Gaborieau B, Vaysset H, Tesson F, et al. Prediction of strain level phage-host interactions across the Escherichia genus using only genomic information. Nat Microbiol. 2024. 9:2847-2861.