VirSorter

VirSorter combines three independent lines of evidence into a unified probabilistic score:

• Viral Hallmark Genes: Detection of capsid, terminase, and other structural proteins using hidden Markov models (HMMs) from Pfam and custom viral databases
• Non-Caudovirales Gene Enrichment: Statistical enrichment of genes characteristic of non-tailed viruses, distinguishing them from prophage elements
• Compositional Departure: Identification of genomic regions with nucleotide composition (k-mer frequencies, GC skew) that deviate significantly from microbial host signatures

This multi-signal approach enables detection of highly divergent viruses that lack detectable nucleotide similarity to known references.

VirSorter assigns predictions into confidence categories to help researchers prioritize results:

• Category 1 (Most Confident): Contigs with strong viral hallmark gene evidence AND significant compositional departure from microbial norms
• Category 2 (Likely): Contigs with viral hallmark genes but weaker compositional signal, or strong compositional signal without hallmark genes
• Category 3 (Possible): Contigs with marginal evidence, often short sequences or prophage regions integrated into host chromosomes
• Category 4-6: Progressively lower confidence, often representing prophages, gene transfer agents, or false positives

This tiered system reduces manual curation burden by surfacing the highest-confidence viral predictions first.

VirSorter includes specialized logic for delineating prophage integration boundaries within microbial genomes:

• Identifies the transition zones where viral-like compositional signatures revert to host-like signatures
• Uses sliding window analysis of k-mer frequencies and gene content to pinpoint attachment sites
• Distinguishes between active prophages (retaining viral gene synteny) and degraded prophage remnants (gene order disrupted by host evolution)

This capability is critical for studying lysogenic conversion, horizontal gene transfer, and the evolutionary arms race between phages and their hosts.

VirSorter relies on a curated hierarchical reference database rather than raw sequence similarity:

• HMM Profiles: Custom-built hidden Markov models trained on viral protein families from RefSeqVirus and manually curated prophage datasets
• Taxonomic Stratification: Separate models for different viral groups (Caudovirales, ssDNA viruses, RNA viruses) to improve specificity
• Host Contamination Filters: Explicit exclusion of HMMs matching conserved bacterial and archaeal proteins to reduce false positives

This profile-based approach captures remote homology that BLAST-based methods miss, enabling detection of viruses with no close sequenced relatives.

VirSorter implements post-hoc filtering to remove non-viral sequences that pass initial scoring:

• Ribosomal RNA Screening: Flags contigs containing 16S/18S rRNA genes, which are definitive markers of cellular organisms
• Universal Single-Copy Gene Check: Cross-references against bacterial/archaeal marker gene sets (e.g., TIGRFAMs) to identify misclassified cellular contigs
• Plasmid Discrimination: Uses replicon typing and conjugation gene detection to separate viral sequences from conjugative plasmids and integrative conjugative elements (ICEs)

This decontamination step is essential for generating high-purity viral genome catalogs from complex metagenomes.

VirSorter outputs are designed for seamless integration into broader metagenomic workflows:

• CheckV Compatibility: Viral contigs can be directly passed to CheckV for completeness estimation, host contamination removal, and quality assessment
• DRAM-v Integration: Predicted viral contigs feed into DRAM-v for functional annotation of auxiliary metabolic genes (AMGs)
• vConTACT2 Clustering: Outputs support protein-sharing network analysis for taxonomic assignment of novel viral clusters
• iVirus Ecosystem: Part of the CyVerse-supported iVirus toolkit, enabling cloud-based analysis of large-scale viromic datasets

This interoperability makes VirSorter a foundational component of modern viral ecogenomics pipelines.