This work reports the method ClassiPhage to classify phage genomes using sequence derived taxonomic features. ClassiPhage uses a set of phage specific Hidden Markov Models (HMMs) generated from clusters of related proteins. The method was validated on all publicly available genomes of phages that are known to infect Vibrionaceae. The phages belong to the well-described phage families of Myoviridae, Podoviridae, Siphoviridae, and Inoviridae. The achieved classification is consistent with the assignments of the International Committee on Taxonomy of Viruses (ICTV), all tested phages were assigned to the corresponding group of the ICTV-database. In addition, 44 out of 58 genomes of Vibrio phages not yet classified could be assigned to a phage family. The remaining 14 genomes may represent phages of new families or subfamilies. Comparative genomics indicates that the ability of the approach to identify and classify phages is correlated to the conserved genomic organization. ClassiPhage classifies phages exclusively based on genome sequence data and can be applied on distinct phage genomes as well as on prophage regions within host genomes. Possible applications include (a) classifying phages from assembled metagenomes; and (b) the identification and classification of integrated prophages and the splitting of phage families into subfamilies.
Evolutionary shifts in bacterial virulence are often associated with a third biological player, for instance temperate phages, that can act as hyperparasites. By integrating as prophages into the bacterial genome they can contribute accessory genes, which can enhance the fitness of their prokaryotic carrier (lysogenic conversion). Hyperparasitic influence in tripartite biotic interactions has so far been largely neglected in empirical host-parasite studies due to their inherent complexity. Here we experimentally address whether bacterial resistance to phages and bacterial harm to eukaryotic hosts is linked using a natural tri-partite system with bacteria of the genus Vibrio, temperate vibriophages and the pipefish Syngnathus typhle. We induced prophages from all bacterial isolates and constructed a three-fold replicated, fully reciprocal 75 × 75 phage-bacteria infection matrix. According to their resistance to phages, bacteria could be grouped into three distinct categories: highly susceptible (HS-bacteria), intermediate susceptible (IS-bacteria), and resistant (R-bacteria). We experimentally challenged pipefish with three selected bacterial isolates from each of the three categories and determined the amount of viable Vibrio counts from infected pipefish and the expression of pipefish immune genes. While the amount of viable Vibrio counts did not differ between bacterial groups, we observed a significant difference in relative gene expression between pipefish infected with phage susceptible and phage resistant bacteria. These findings suggest that bacteria with a phage-susceptible phenotype are more harmful against a eukaryotic host, and support the importance of hyperparasitism and the need for an integrative view across more than two levels when studying host-parasite evolution.
Background/ Motivation: In the era of affordable next generation sequencing technologies we are facing an exploding amount of new phage genome sequences. This requests high throughput phage classification tools that meet the standards of the International Committee on Taxonomy of Viruses (ICTV). However, an accurate prediction of phage taxonomic classification derived from phage sequences still poses a challenge due to the lack of performant taxonomic markers. Since machine learning methods have proved to be efficient for the classification of biological data we investigated how artificial neural networks perform on the task of phage taxonomy. Results: In this work, 5,920 constructed and refined profile Hidden Markov Models (HMMs), derived from 8,721 phage sequences classified into 12 well known phage families, were used to scan phage proteome datasets. The resulting Phage Family-proteome to Phage-derived-HMMs scoring matrix was used to develop and train an Artificial Neural Network (ANN) to find patterns for phage classification into one of the phage families. Results show that using the 100 fold cross-validation test, the proposed method achieved an overall accuracy of 84.18 %. The ANN was tested on a set of unclassified phages and resulted in a taxonomic prediction. The ANN prediction was benchmarked against the prediction resulting of multi-HMM hits, and showed that the ANN performance is dependent on the quality of the input matrix. Conclusions: We believe that, as long as some phage families on public databases are underrepresented, multi-HMM hits can be used as a classification method to populate those phage families, which in turn will improve the performance and accuracy of the ANN. We believe that the proposed method is an effective and promising method for phage classification. The good performance of the ANN and HMM based predictor indicates the efficiency of the method for phage classification, where we foresee its improvement with an increasing number of sequenced viral genomes.