Phages are often highly specific to their bacterial hosts, a trait that opens avenues for targeted applications in phage therapy. Accurately identifying which phages can infect specific bacterial strains could speed up the design of effective phage-based treatments. In a recent study, Gaborieau, Vaysset, Tesson et al. developed a method to predict phage–host specificity using genomic data from a diverse set of Escherichia strains.

The researchers assembled two collections: the Picard collection of 403 Escherichia strains and the Guelin collection of 96 phages, both of which are genomically characterized. They used plaque assays to experimentally establish a dataset of 38,688 phage–bacteria interactions, which was then used to train predictive algorithms aimed at identifying genomic traits that determine phage–host specificity. Adsorption factors were crucial for predicting phage–bacteria interactions within the Escherichia genus, whereas antiphage systems played a minor role. The authors also developed a machine learning approach to design phage cocktails targeting specific pathogenic Escherichia coli strains using only bacterial genomic data. The tailored cocktails were more effective at killing bacteria than generic or standard combination cocktails, highlighting the potential of using genomic data to design phage cocktails.