Phylogeny-agnostic strain-level prediction of phage-host interactions from genomes using machine learning.

A machine-learning framework predicts strain-level phage-host interactions across diverse bacteria with AUROC up to 0.94, independent of phylogenetic constraints.

• Why it matters: Selecting effective phages for targeting specific bacterial strains is a major challenge limiting phage therapy and microbiome engineering. Addressing this gap can improve precision in these applications.
• What they did: The authors developed and optimized a phylogeny-agnostic machine-learning approach using 13.2 million training runs on six datasets, involving 115,037 interactions, 949 bacterial strains, and 518 phages.
• The result: Experimental validation confirmed the model's generalizability with AUROC 0.84, enabling up to 97.5% bacterial coverage with five phages and improving single-phage selection efficiency, supporting rational phage therapy design.