Researchers have introduced metaIVP, an open-source computational framework designed to purify, evaluate and interpret viral and non-viral genomic content from metagenomic sequencing data within a single, reproducible workflow. Described in BMC Methods, the pipeline addresses a persistent gap in virome analysis: while many existing tools excel at recovering viral genomes, they often leave the non-viral fraction of a sample uncharacterized and provide limited support for selective re-analysis when results need to be traced or revised.
Metagenomic studies increasingly depend on complex, multi-tool pipelines to recover and characterize genomes from mixed microbial communities. In metavirome work, where viral and non-viral genomes must be handled with distinct methodologies, the authors note that limited downstream post-processing functionality and insufficient logging structures can hinder reproducibility, error tracing and selective re-analysis. Experimental enrichment of virus-like particles, traditionally achieved through differential centrifugation, size-based filtration and density gradient separation, involves tradeoffs between enrichment efficiency and discovery potential, because assumptions about particle size and density can bias against atypical or large viruses. Moreover, experimental purification rarely achieves complete viral purity, and substantial non-viral genomic material often remains in metaviromic sequencing data, motivating complementary computational strategies.
metaIVP is organized into hierarchical modules, each governed by dedicated log files that explicitly control execution state and re-runnability. Contig-level and bin-level purification steps isolate genome contents, after which separate viral and non-viral post-processing workflows run. Viral workflows incorporate contamination detection, genome quality evaluation, host prediction and virus-specific binning, while non-viral analyses include genome binning, alignment and mapping statistics, genome quality assessment and replication rate estimation. Checkpoints are defined so that deleting selected module- or sub-module-level logs enables targeted re-execution of specific analytical steps without rerunning the entire pipeline. For example, re-executing viral re-binning with vRhyme requires removal of the corresponding purification module log, the viral post-processing control log and the vRhyme-specific process log. The pipeline is implemented with Linux shell scripting and Python-based orchestration and does not depend on external workflow languages or additional bioinformatic infrastructure.
At the contig level, metaIVP applies a flexible two-step viral recovery strategy. The pipeline first uses CheckV to assess viral signals, quality and completeness for individual contigs, enabling confident identification and purification of viral contigs. Contigs that cannot be confidently classified by CheckV are not discarded; instead, they are further analyzed with geNomad to recover additional viral sequences based on complementary viral signatures. This allows metaIVP to systematically recover viral content that may fall outside the strict classification thresholds of a single tool. Following contig-level classification, the pipeline constructs bin-level sequence files by aggregating contigs according to existing bin assignments, then re-evaluates these bins with CheckV for completeness and contamination. High-confidence viral bins are retained for downstream viral-focused analyses, while undetermined bins and residual contigs are collected into a non-viral genome set, enabling parallel characterization of both components from the same dataset.
Optional re-binning steps refine genome reconstruction for both fractions. Because re-binning can substantially increase computational time and memory usage, particularly for large datasets, these steps are not required for routine execution and are recommended for Linux servers or high-performance computing environments. For viral genomes, metaIVP supports re-binning with vRhyme, a state-of-the-art viral binning method with performance reported to be comparable to the widely used MVP pipeline; re-binned viral genomes are re-evaluated with CheckV and geNomad. For non-viral content, MetaBAT2 can re-bin residual contigs classified as non-viral or undetermined, with the resulting microbial bins assessed by CheckM2 for completeness and contamination. Downstream, geNomad generates refined viral annotations, iPHoP infers potential virus-host associations with confidence scores, and iRep estimates bacterial replication rates from the non-viral genomes. Host predictions in the study were filtered at an iPHoP score of at least 0.7, a cutoff the authors adopted to maintain sensitivity while controlling error rates, and users can adjust this threshold.
The authors demonstrated metaIVP on two publicly available datasets representing different levels of complexity. The first, the Copenhagen dataset, consists of six human gut virome samples, three of which were washed three times before sequencing and three of which were not; it serves as a viral benchmark for evaluating purification efficacy. The second, the Florida Lake dataset, comprises 20 metagenomic samples of sediment microbial communities from Loxahatchee National Wildlife Refuge, grouped by geographic collection location into North, South, East and West categories, and was used to assess performance on complex environmental samples containing both viral and non-viral genomes. As a pre-analysis step, all samples were processed with a standard assembly and binning pipeline using MEGAHIT and MetaBAT2; in the Copenhagen dataset, more than 80 percent of reads mapped back to assembled contigs, indicating strong data quality.
In the human gut data, metaIVP recovered additional high-quality viral bins after purification, particularly those classified as high quality by amino-acid-identity-based criteria. To quantify bin quality, the pipeline computes a viral bin quality score defined as the fraction of viral genes relative to the total number of viral and host genes as annotated by CheckV. For bins classified as high quality, this score consistently increased after applying metaIVP, indicating enrichment of viral genomic content and reduction of residual host contamination. On average, high-quality bins showed a median improvement of 0.048 and a mean improvement of 0.101 across 32 bins. Paired before-and-after comparisons revealed a clear positive shift in viral gene fraction, which the authors interpret as evidence that the pipeline improves bin purity rather than merely redistributing contigs among bins. Enrichment was also observed across the full bin set, including medium- and low-quality bins.
Host prediction in the gut dataset proved robust. Retaining iPHoP predictions with scores of at least 0.7, the researchers found that host predictions were highly consistent across all six samples. At the phylum level, Bacteroidota and Bacillota_A dominated the predicted hosts, with Bacteroidota-associated hosts detected in all six samples, indicating a stable and reproducible host-virus signal in this cohort. Additional host phyla, including Actinomycetota and Bacillota, appeared in multiple samples, often in two or more independent libraries despite differences in storage and washing protocols. Analysis of bacterial content using CheckM2 and iRep indicated that the SM buffer storage condition retained the fewest detectable bacterial genomes and contigs, consistent with previously reported findings for this dataset.
In the Florida environmental samples, principal component analysis of enzyme commission abundance profiles from the initial metagenomic assemblies showed clear separation of samples by geographic location, indicating strong location-associated functional signals at the metagenomic level. After metaIVP enriched for viral genomes, the pipeline yielded a higher fraction of high-quality viral bins defined by the AAI-High criterion, although a fraction of HMM-High bins was lost. Across the 20 samples, metaIVP removed a substantial fraction of assembled contigs during virome refinement, with fewer than approximately one-third of the original contigs contributing to the final viral genome set. While alternative approaches such as MVP and vRhyme produced a larger absolute number of high-quality bins, they also generated substantially more lower-quality and unclassified bins, resulting in a lower overall proportion of high-confidence viral genomes. After purification, EC-level quantifications from the refined virome bins no longer exhibited strong location-driven separation, with samples from different locations showing increased overlap, suggesting that the host-associated or environmental bacterial signals driving the original separation were substantially reduced.
Host prediction across the environmental viromes showed highly consistent score distributions across all four geographic locations, with no clear location-specific shifts, indicating that predicted virus-host association quality is robust across spatially separated environments. Several host phyla, including Desulfobacterota, Gemmatimonadota, Thermoplasmatota, Halobacterota, Acidobacteriota, Pseudomonadota, Actinomycetota and Nitrospirota, showed iPHoP scores above 0.95 and were detected in all four locations, suggesting broadly distributed viral-host relationships. Other groups, such as Methanobacteriota and Patescibacteria, appeared only in a single location, pointing to localized or low-abundance host populations. For the retained non-viral genomes, contamination and completeness showed broadly overlapping distributions across directions, and relationships between replication rate and genome quality showed substantial genome-to-genome variability without consistent global trends. One exception was a modest shift toward higher iRep replication rates in samples from the South, which the authors caution may reflect elevated microbial activity or distinct environmental conditions but may also result from sampling depth, assembly quality or coverage-related effects, requiring further targeted analysis to distinguish biological signal from technical artifacts.
The authors compared metaIVP against the MVP pipeline, run on the same datasets with default parameters and the same assembled contigs as input, with MVP’s viral bins re-evaluated using CheckV and geNomad under the same criteria. They report that metaIVP achieves comparable performance, particularly when optional re-binning with vRhyme is applied, while maintaining a higher fraction of high-confidence viral bins. The team also acknowledges limitations: the study did not include comprehensive simulated or mock community benchmarking, because constructing realistic metavirome simulations remains challenging and dataset dependent, and the observed improvements in viral bin purity for high-quality bins are modest in the well-assembled, low-diversity datasets shown and may vary with data complexity. Future extensions could incorporate additional tools such as VirSorter2 and language model-based approaches like ViraLM to enable consensus-based strategies. metaIVP is available as an open-source Python package on GitHub, with example scripts, input and output data, and the files and notebook used to generate the paper’s figures and tables deposited on Zenodo.
Subject of Research: A modular computational pipeline for purifying and analyzing viral and non-viral genomes in metavirome data
Article Title: metaIVP: an integrative metavirome focused metagenomic processing pipeline
Article References: Sahu, K., & Yao, Q. (2026). metaIVP: an integrative metavirome focused metagenomic processing pipeline. BMC Methods, 3(1), Article 37. https://doi.org/10.1186/s44330-026-00090-7
Image Credits: AI Generated
DOI: 10.1186/s44330-026-00090-7
Keywords: metavirome, metagenomics, viral binning, CheckV, geNomad, iPHoP, vRhyme, host prediction, virus-host interactions, pipeline reproducibility, human gut virome, environmental viromes
News Source: Kristina Jarvis. (October 28, 2026). New metaIVP pipeline refines viral and non-viral genomes from metagenomes. Scienmag.
