AI model predicts which phages infect which E. coli strains from the bacterial genome alone (Nature Microbiology)

Scientists from the Institut Pasteur, Inserm, AP-HP and Université Paris Cité trained an AI model that recommends a phage cocktail for a bacterial strain from its genome alone, using 350,000 interactions between 403 E. coli strains and 96 phages. The model correctly predicted phage efficacy in 85% of cases in the dataset, and in 90% of cases the phages it chose for a new collection of pneumonia-causing E. coli strains destroyed the bacteria; infection depended on receptors at the bacterial surface rather than on anti-viral defence mechanisms. The work was published on October 31, 2024 in Nature Microbiology.