What it is
The authors present a machine-learning framework that predicts which phages infect which bacterial strains, across diverse bacterial genera, from genome sequences alone. Optimized over 13.2 million training runs on six datasets (115,037 interactions, 949 bacterial strains, 518 phages), it matched species-specific methods (AUROC 0.67 to 0.94) without their phylogenetic constraints. Tested experimentally on 1,240 predicted E. coli phage-host interactions it reached an AUROC of 0.84, genome-wide RB-TnSeq screens showed it captured 68.6% of experimentally identified infection mediators, and model-guided cocktails reached up to 97.5% bacterial coverage with five phages.
Why it matters
Phages are promising alternatives to antibiotics for drug-resistant infections and tools for engineering microbiomes, but selecting phages that infect a specific bacterial strain has limited their use. A predictor that works from genomes, without being tied to one species' phylogeny, is presented by the authors as enabling rational phage-therapy design and precision microbiome engineering, with applications across clinical, agricultural and industrial settings.
Every metric behind this entry is listed, with its source, under Sources and data below.
Filed underBacteriophages and microbial interactions, Genomics and Phylogenetic Studies, Machine Learning in Bioinformatics