henhao
Sign in
BioinformaticsbioRxiv
Preprint

CheckV: assessing the quality of metagenome-assembled viral genomes

Stephen NayfachiD, Antônio Pedro CamargoiD, Emiley A. Eloe‐FadroshiD, Simon RouxiD, Nikos C. KyrpidesiD

Abstract

Over the last several years, metagenomics has enabled the assembly of millions of new viral sequences that have vastly expanded our knowledge of Earths viral diversity. However, these sequences range from small fragments to complete genomes and no tools currently exist for estimating their quality. To address this problem, we developed CheckV, which is an automated pipeline for estimating the completeness of viral genomes as well as the identification and removal of non-viral regions found on integrated proviruses. After validating the approach on mock datasets, CheckV was applied to large and diverse viral genome collections, including IMG/VR and the Global Ocean Virome, revealing that the majority of viral sequences were small fragments, with just 3.6% classified as high-quality (i.e. > 90% completeness) or complete genomes. Additionally, we found that removal of host contamination significantly improved identification of auxiliary metabolic genes and interpretation of viral-encoded functions. We expect CheckV will be broadly useful for all researchers studying and reporting viral genomes assembled from metagenomes. CheckV is freely available at: http://bitbucket.org/berkeleylab/CheckV.

PDFDOICodeData
Full-text digest ↓

02 /Specification & Classification

Click any tag to browse matching papers
Citations
61
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (8 May 2020) · CC BY 4.0

CheckV estimates completeness of metagenome-assembled viral contigs against GenBank plus 76,262 metagenome-derived complete genomes and trims host flanks; 86.9% of IMG/VR contigs were low-quality.

Metagenomic viral contigs range from short fragments to complete genomes and may carry host DNA from integrated proviruses, but viruses lack the widely distributed single-copy markers that CheckM relies on, so studies typically apply a uniform length cutoff of 5 or 10 kb. CheckV estimates completeness for each contig (VIBRANT only assigns quality tiers) and locates host-virus boundaries by comparing gene content and GC content in sliding windows.

Key results

  1. ShownOn mock fragments of 382 phages that were not in the reference database, CheckV's completeness estimates closely tracked true completeness. Accuracy dropped moderately when only distant references were available.

    Median unsigned error was 0.91% with the full database. It rose to 5.5% when restricted to highly diverged references (<30% AAI), while still giving estimates for >96% of fragments. R2 against true completeness was 0.96 for CheckV estimates, 0.78 for CheckV quality tiers and 0.35 for VIBRANT quality tiers. · Accurate estimation of completeness and…; Benchmarking CheckV and comparison to existing…

  2. ShownOn mock proviruses, CheckV detected host DNA flanking proviruses more often than four existing provirus tools, and it rarely flagged purely viral fragments.

    CheckV flagged 76.5% of mock proviruses vs 20.7 to 33.5% for VIBRANT, VirSorter, PhiSpy and Phigaro. It misclassified 0.9% of entirely viral fragments; PhiSpy called non-viral regions on 22.9% of them (covering 26.3% of their length on average). VirSorter detected 1.5% of proviruses with <20% contamination. A provirus call required a viral region covering <95% of the fragment. · Accurate estimation of completeness and…; Benchmarking CheckV and comparison to existing…

  3. ShownApplied to 735,106 IMG/VR 2.0 contigs, CheckV classified most as low-quality fragments. Direct terminal repeats (DTRs) agreed with high estimated completeness far more often than inverted terminal repeats (ITRs) or two-sided provirus flanks.

    Tiers: 1.9% complete, 2.5% high-, 6.5% medium-, 86.9% low-quality, 2.2% undetermined. AAI-based estimates covered 572,369 contigs (78.6%) at high or medium confidence; adding the HMM approach raised coverage to 97.8%. Among putatively complete contigs with an estimate, 90.1% of DTR contigs vs 58.2% of complete proviruses and 31.8% of ITR contigs were >90% complete. · Using CheckV to identify high-quality genomes…; Application of CheckV to the IMG/VR and GOV…

  4. ShownHost-derived flanks were uncommon in IMG/VR but held a disproportionate share of metabolic gene annotations, so without trimming, host functions can be mistaken for viral-encoded ones.

    5.2% of IMG/VR contigs had host flanks, 97.1% of them on one side only (median 12.0% of contig length). Host regions held 12.0% of provirus genes but 42.3% of genes in KEGG metabolic pathways. Of 625 multidrug-resistance efflux pump genes, 92.8% lay in host regions. 617 of 805 long proviruses (>50 kb, >20% contamination; 76.6%) were also called by VirSorter or VIBRANT. · Using CheckV to discriminate viral-encoded…; Application of CheckV to the IMG/VR and GOV…

  5. ShownSearching public metagenomes, metatranscriptomes and viromes for DTR contigs greatly widened the reference set of complete viral genomes beyond GenBank, especially for Caudovirales.

    Searching 14.4 billion contigs (9.7 Tb) yielded 76,262 filtered DTR contigs (39,117 at 95% ANI). Merged with 24,834 GenBank genomes, they formed 52,141 clusters, a 3.9-fold increase in clusters (7.1-fold for Caudovirales). GenBank still gave better coverage of RNA viruses (Riboviria). · An expanded database of complete viral genomes…; Database of complete viral genomes for AAI-based…; Validating the completeness of CheckV reference…; Generating a non-redundant set of CheckV…

Methods, data and reproducibility

CheckV labels Prodigal genes as viral or microbial with 15,958 curated HMMs and scores each intergenic position by gene-label and GC differences between flanking windows (up to 40 genes) to trim host regions. Completeness is contig length over an expected genome length from DIAMOND AAI hits to complete genomes (conservative HMM fallback); DTRs, ITRs and two-sided host flanks mark closed genomes.

  • NCBI GenBank viral genomes (downloaded June 1, 2019), filtered for completeness to seed the reference database · viruses · n = 30,903 downloaded; 24,834 retained
  • Assembled contigs from public metagenomes, metatranscriptomes and viromes (IMG/M, MGnify, Pasolli et al., HGM, HuVirDB, global ocean viromes) searched for DTR contigs · viruses (environmental and human-associated) · n = 14.4 billion contigs searched; 76,262 DTR contigs retained
  • GTDB Release 89 genomes (one per family) and GenBank viral proteins, used to select viral- vs microbial-specific HMMs · bacteria, archaea, viruses · n = 2,015 bacterial and 239 archaeal genomes; 1,590,764 viral proteins
  • Mock benchmark: GenBank phage genomes (after June 1, 2019) paired by genus with GTDB host genomes to simulate genome fragments and proviruses · bacteriophages; bacterial and archaeal hosts · n = 382 phages; 76 hosts (71 bacterial, 5 archaeal); 10 fragments per genome
  • IMG/VR 2.0 viral contigs >5 kb (cultivated isolates and proviruses from microbial genomes excluded) · viruses · n = 735,106 contigs
  • Global Ocean Virome (GOV) 2.0 viral contigs >5 kb or circular · viruses (open-ocean viromes) · n = 488,131 contigs
  • Largest DTR contig found in IMG/VR, from an Antarctic saline lake, annotated as a putative huge phage via TerL phylogeny · phage · n = 1 contig (528,258 bp) · IMG ID 3300025697_Ga0208769_1000001
Methods & tools
CheckV, Prodigal v2.6.3, HMMER hmmsearch v3.1b2, DIAMOND, BLAST+ v2.9.0 (blastn, dustmasker), VirFinder v1.1, VIBRANT v1.2.0, VirSorter v1.0.5, PhiSpy v3.7.8, Phigaro v0.1.5.0, Prokka v1.14.5, HMM sources: VOGDB, IMG/VR, RVDB, KEGG Orthology, Pfam, TIGRFAM
Code
https://bitbucket.org/berkeleylab/CheckV
Data & accessions
https://datacommons.cyverse.org/browse/iplant/home/shared/iVirus/GOV2.0http://vogdb.org/
Software & environment
Written in Python; BSD licence. Stated versions: Prodigal v2.6.3, hmmsearch v3.1b2, BLAST+ v2.9.0, VirFinder v1.1 and the comparison tools; DIAMOND version not given; no container or requirements file described.
Assay conditions & compute
On 16 CPUs (Intel Xeon E5-2698 v3), CheckV ran 1.6x to 11.6x faster than the compared tools on the mock dataset, using about 2 GB RAM. On a single CPU it was faster than VirSorter and VIBRANT but slower than PhiSpy and Phigaro.
Reproducibility gaps
CheckV quality statistics and the list of complete genomes for IMG/VR are deferred to a later IMG/VR release (scheduled for September 2020); no separate download location is given for the reference genome and HMM databases.
Limitations (6)
  • The reference database needs ongoing additions of new viral genomes to widen its environmental and taxonomic coverage. · stated by the authors
  • Viral bins and segmented genomes that span several sequences are not handled by this version. · stated by the authors
  • Benchmarks and applications centred on bacterial and archaeal viruses; extending CheckV to eukaryotic viruses, notably segmented ones or those rich in metabolic genes, may require changes. · stated by the authors
  • Read mapping is not yet used; the authors propose it to detect circular contigs, refine host-virus boundaries and locate genome termini. · stated by the authors
  • Accuracy comes only from author-built mock data (382 GenBank phages, hosts paired by genus). Genomes used to train CheckV were excluded from the benchmark, but boundary parameters were tuned on 'a dataset of mock proviruses' and IMG/VR, and the text does not say how that set relates to the benchmark proviruses. · noted in this summary
  • IMG/VR 2.0 and GOV 2.0 supplied DTR contigs to the reference database and were then scored against it, with only perfect matches (100% AAI and 100% AF) excluded. Near-identical relatives may make completeness estimates on these collections look better than they would on new data. · noted in this summary

Who should read:Read this if you assemble or curate viral contigs from metagenomes and need completeness estimates for each contig, trimming of host regions from proviruses, or a large reference set of complete environmental viral genomes.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISLi '24Sutton '19Chen '23Zolfo '24Jochheim '24Rodrigues '26
68% semantic matchBioinformatics
Bowen Li, Xianyue Jiao, Guanxiang Liang·bioRxiv·Jan 21, 2024· 1 citations