Particle.news
Get it on Google Play
Download on the App Store
29 ARTICLES
·
last mo.

AI Designs 16 Novel Bacteriophages That Infect E. coli

The result points to new, customizable phage therapies alongside an urgent need to strengthen screening and oversight for generative genomics.

Image
Two people working together at a desk in what appears to be a laboratory or research office environment. Large computer monitors display colorful three-dimensional molecular or protein structure models against dark backgrounds, while a laptop on the desk shows lines of code or data. One person is seated using the laptop, and another is leaning toward the screens as both focus on the scientific visualizations. Laboratory shelving and equipment are visible in the background, suggesting a biomedical, biotechnology, or computational research setting. The scene emphasizes the use of computer-based analysis and visualization tools to study complex biological structures.
Image
Image

Overview

  • A Stanford- and Arc Institute-led team used genome‑language models called Evo, Evo1 and Evo2 to generate complete viral genomes and published the peer-reviewed results in Science on Thursday.
  • The models scanned trillions of nucleotides and produced roughly 700,000 candidate genomes, of which researchers synthesized about 285 and found 16 produced viable bacteriophages that infect Escherichia coli.
  • The AI-designed phages cannot infect humans because the team excluded human‑pathogen data from model training and chose the ΦX174 family, which targets only E. coli.
  • Lab tests showed a cocktail of the new phages overcame E. coli resistance that natural ΦX174 could not, indicating possible routes to treat antibiotic‑resistant infections.
  • Biosecurity experts and a Johns Hopkins commentary warned that the capability exposes gaps in DNA‑order screening, oversight and detection for AI‑generated genomes, a debate intensified by the public release of Evo2.