ViromeScan: a new tool for metagenomic viral community profiling

Abstract

Background: Bioinformatics tools available for metagenomic sequencing analysis are principally devoted to the identification of microorganisms populating an ecological niche, but they usually do not consider viruses. Only some software have been designed to profile the viral sequences, however they are not efficient in the characterization of viruses in the context of complex communities, like the intestinal microbiota, containing bacteria, archeabacteria, eukaryotic microorganisms and viruses. In any case, a comprehensive description of the host-microbiota interactions can not ignore the profile of eukaryotic viruses within the virome, as viruses are definitely critical for the regulation of the host immunophenotype.

Results: ViromeScan is an innovative metagenomic analysis tool that characterizes the taxonomy of the virome directly from raw data of next-generation sequencing. The tool uses hierarchical databases for eukaryotic viruses to unambiguously assign reads to viral species more accurately and >1000 fold faster than other existing approaches. We validated ViromeScan on synthetic microbial communities and applied it on metagenomic samples of the Human Microbiome Project, providing a sensitive eukaryotic virome profiling of different human body sites.

Conclusions: ViromeScan allows the user to explore and taxonomically characterize the virome from metagenomic reads, efficiently denoising samples from reads of other microorganisms. This implies that users can fully characterize the microbiome, including bacteria and viruses, by shotgun metagenomic sequencing followed by different bioinformatic pipelines.

Fig. 1

Workflow of ViromeScan. a Inputs are single-end reads (fastq format) or paired-end reads (fastq or compressed fastq format). b Candidate viral reads are identified by mapping the sequences to the selected reference database. Unmapped reads are not contained in the resulting file. c Three filtering procedures to trim low quality reads and completely remove human and bacterial contaminations are computed. d The remaining viral sequences are assigned to appropriate taxonomy and the results are tabulated as both relative abundance and read counts

Fig. 2

Comparison of ViromeScan to other existing methods. A total of five synthetic viral communities were used in order to compare ViromeScan with Metavir [22] and blastN [29]. Absolute and r.m.s. errors in assigning taxonomy at (a) family and (b) species level are shown. c Correlation between predicted and real relative abundance for the 5 non-evenly distributed mock communities. d Read rate for the tested tools on single CPU

Fig. 3

Comparison between the relative abundances of a single non-evenly distributed mock community as detected using Metavir [22], blastN [29] and ViromeScan, and its real composition. Black portions of the bars correspond to the unassigned viral fraction or erroneous viral assignment

Fig. 4

Different body sites reflect different virome configurations. Species (a) and families (b) level hierarchical Ward-linkage clustering based on the Spearman correlation coefficients of the viral profiles of 20 HMP samples [27] as determined by ViromeScan. Analysis was carried out considering all the families detected and species with at least 0.5 % of abundance in 25 % of samples. c Hierarchical Ward-linkage clustering of viral families generated characteristic clades, which discriminated the gut environment from the other body sites. The names of the families are colored according to the colors of the dendrogram (b)

Fig. 5

The eukaryotic virome at family level in an asymptomatic Western population, as predicted by ViromeScan. Analysis was carried out on 20 HMP samples [27] from 4 human body sites, including gut (stool), mouth (buccal mucosa), skin (retroauricular crease) and vagina (mid vagina). a The relative abundance of viral families for each HMP sample and the mean relative abundance for each body site are reported in the histograms and pie charts, respectively. b Hierarchical Ward-linkage clustering based on the Spearman correlation coefficients of 19/20 HMP samples containing members of the human Herpesviridae family