Current approaches for bacteriophage-host prediction are strongly biased toward a few cultured genera. Here, we will develop methods for systematically and reliably identifying phage-host relationships in human gut microbiomes, including non-temperate phages.

First, we will develop a statistical method that integrates the scores from five orthogonal sources of information while not requiring any labeled data. Instead, the method will produce increasingly reliable labels (i.e., predicted phage-host relations) during its iterative, semisupervised training.

Second, we will develop statistical graphical models to predict phage-host relationships from population-level high-throughput sequencing data. We consider both co-abundance data from marker-gene/metagenomics sequencing and metatranscriptome data of phage mRNAs and the transcriptional signature of hosts to acute lytic infection.

Third, we will develop a viral tagging method to screen metagenomic libraries of phages against metatranscriptomic bacterial libraries to detect which phages infect which hosts. Single bacterial cells plus associated phages will be analysed using pair-of-single-host-bacteria-single-phage (POSH) RNA-seq, where phage plus host will be identified by their mRNAs.