ViroSeek: a viral detection pipeline for second-generation sequencing

This paper introduces ViroSeek, a lightweight, reproducible, and accessible bioinformatics pipeline designed to simplify viral taxonomic analysis of second-generation sequencing data through an automated workflow that has been empirically validated for accurate virus detection and contaminant removal.

The Gist

Imagine you are a detective trying to solve a mystery in a bustling, chaotic city. This city is a drop of mosquito blood, and the "criminals" you are looking for are tiny, invisible viruses.

In the past, finding these criminals was like searching for a specific needle in a haystack by looking at one piece of hay at a time. It was slow, expensive, and you could only find one type of needle.

Now, we have a new tool called NGS (Next-Generation Sequencing). Think of this as a super-fast camera that takes a picture of every single piece of hay in the entire city at once. Suddenly, you have a mountain of data containing the DNA of the viruses, the mosquito itself, bacteria, and all sorts of junk.

The Problem:
The mountain of data is too messy. It's like trying to find a specific suspect in a crowd of 10,000 people where 9,900 are innocent bystanders (the mosquito's own DNA) and 99 are just random tourists (bacteria). Existing tools to sort this out are often:

• Too complicated: Like a spaceship control panel that only a PhD astronaut can fly.
• Broken: Like a map that leads you in circles.
• Wrong: They might tell you the suspect is a "human" when they are actually a "robot" (misidentifying the virus).

The Solution: ViroSeek
The authors of this paper built ViroSeek. Think of ViroSeek as a smart, automated sorting machine designed specifically for this job. It's lightweight, easy to use, and doesn't require a PhD to operate.

Here is how ViroSeek works, step-by-step, using our detective analogy:

1. The Cleanup Crew (Pre-processing):
   First, ViroSeek sweeps the floor. It throws away blurry photos (bad quality data) and cuts off the sticky tape (adapters) that got stuck on the DNA during the lab process. It's like cleaning your glasses before you try to read a map.

2. The Bouncer (Host Removal):
   The machine has a VIP list. It knows what the "mosquito" looks like and what "bacteria" looks like. It kicks them out of the club immediately. Now, the room is much quieter, and we only have the "suspects" (viruses) left.

3. The Puzzle Solver (Assembly):
   The DNA of the viruses is broken into tiny, scattered puzzle pieces. ViroSeek takes these pieces and tries to put them back together to see the full picture of the virus. It's like assembling a jigsaw puzzle where the pieces are scattered on the floor.

4. The ID Check (Taxonomic Assignment):
   Once the puzzle is assembled, ViroSeek holds the picture up to a giant "Wanted Poster" database (a library of known viruses). It asks, "Who are you?" It uses a super-fast matching system (DIAMOND) to find the best match.

   • Crucial Detail: Unlike other tools that might guess, ViroSeek checks the "frames" of the puzzle carefully to make sure it doesn't misidentify a virus just because it looks slightly similar to another.

5. The Crowd Counter (Quantification):
   Finally, it counts how many pieces belong to each virus. This tells the researchers not just which viruses are there, but how many of them are present. It also removes "double-counts" (PCR duplicates), ensuring the count is accurate, like making sure you don't count the same person twice in a lineup.

Did it work? (The Results)
The authors tested ViroSeek in a "lab crime scene." They took mosquitoes, infected them with 6 known viruses, and mixed in a lot of mosquito DNA to make it hard.

• The Result: ViroSeek found 100% of the viruses they planted, even when they were very rare.
• The Comparison: They pitted ViroSeek against other famous detective tools (Taxprofiler, MetaDenovo, VirusTaxo).
  • Speed: ViroSeek was 4 to 20 times faster than the others. It solved the mystery in minutes while the others took hours.
  • Accuracy: The other tools often missed viruses or got confused about which virus was which. ViroSeek was sharp and precise.
  • Memory: It didn't need a supercomputer to run; it fit on a standard laptop.

The Catch (Limitations)
Even the best detective makes mistakes if the "Wanted Posters" in the library are wrong.

• Sometimes, two viruses look so similar (like identical twins) that ViroSeek might say, "This is Twin A," when it's actually Twin B. This isn't ViroSeek's fault; it's because the database needs updating.
• If the lab team accidentally mixed up samples (cross-contamination), ViroSeek will faithfully report the wrong virus. It's a mirror; it shows what is there, but it can't fix a messy crime scene before it starts.

The Bottom Line
ViroSeek is a game-changer. It turns a complex, expensive, and confusing scientific process into a simple, reliable, and fast workflow. It allows scientists to quickly identify emerging viruses (like new strains of Dengue or Zika) without needing to be a computer wizard. It's the difference between manually searching a library for a book and using a high-speed barcode scanner that finds it instantly.

Technical Summary

1. Problem Statement

The emergence of arboviruses and zoonotic viruses is a critical public health challenge exacerbated by climate change and globalization. While Next-Generation Sequencing (NGS) offers a high-throughput alternative to traditional culture-based methods for viral detection, existing bioinformatics pipelines for virome analysis suffer from significant limitations:

• Complexity and Accessibility: Many tools are technically complex, difficult to install due to obsolete dependencies, or contain blocking errors, making them inaccessible to non-specialists.
• Specialization Mismatch: Tools designed for third-generation sequencing (e.g., Nanopore) often fail to account for features specific to second-generation sequencing (e.g., PCR duplicates), leading to biased abundance estimates.
• Target Limitations: Many pipelines are optimized for bacteriophages rather than arboviruses or lack the ability to perform per-sample assembly, which is crucial for downstream phylogenetic analysis.
• Usability Issues: Existing solutions like Taxprofiler, MetaDenovo, and VirusTaxo were found to have installation hurdles, unsuitable output formats, or high resource requirements that hindered their adoption in specific research contexts.

2. Methodology

ViroSeek is a lightweight, reproducible, and open-source bioinformatics pipeline designed specifically for the taxonomic analysis of target-enriched second-generation sequencing (Illumina) data.

• Workflow Architecture:

  • Orchestration: Built using Nextflow to ensure portability and reproducibility.
  • Containerization: All tools are encapsulated within Docker or Singularity containers to eliminate dependency conflicts.
  • Input: Paired-end or single-end FASTQ files.
  • Output: A clear, aggregated viral taxonomy table and assembled contigs.

• Processing Steps:

  1. Pre-processing: Quality control (FastQC), adapter trimming, and quality filtering using TrimGalore or fastp.
  2. Contaminant Removal: Filtering of ribosomal RNA and host/bacterial sequences using BBduk against the SILVA rRNA database.
  3. Assembly: De novo assembly of reads per sample using SPAdes (v4.2.0) with the --rnaviral option. Contigs are filtered by a user-defined minimum length (default 140 bp) to reduce false assignments.
  4. Taxonomic Assignment:
     • Alignment of contigs against a protein database using DIAMOND in blastx mode (frameshift-aware).
     • Classification using TaxonKit to map hits to the NCBI taxonomy.
     • Parameters include strict thresholds (e.g., E-value ≤ 1e-5, identity ≥ 96%, query coverage ≥ 51%) to balance sensitivity and specificity.
  5. Quantification: Cleaned reads are remapped to assembled contigs using Minimap2. Samtools is used to mark and remove PCR duplicates, ensuring accurate relative abundance estimation.

• Validation Strategy:

  • Experimental Samples: Four pools of Aedes albopictus mosquitoes experimentally infected with six known viruses (Alphaviruses, Orthoflaviviruses, Orthobunyaviruses) at varying dilutions (MixA, MixB, MixC) and a co-infection sample (AltMix) containing a DNA virus (Aedes albopictus densovirus 2).
  • Simulated Data: A metagenomic dataset generated via CAMISIM containing Hepatitis A virus, Ippy virus, and a bacterial contaminant (Helicobacter hepaticus).
  • Comparative Analysis: ViroSeek was benchmarked against Taxprofiler, MetaDenovo, and VirusTaxo regarding computational efficiency (CPU time, RAM) and detection sensitivity.

3. Key Contributions

• Specialized Design for Second-Generation Data: Unlike many modern tools, ViroSeek explicitly handles PCR duplicates and is optimized for short-read Illumina data, preventing abundance bias.
• Per-Sample Assembly: It retains assemblies on a per-sample basis, enabling downstream phylogenetic placement and genetic diversity studies, a feature missing in some aggregate-assembly pipelines.
• Accessibility: The pipeline is fully documented, open-source (GitHub), and designed to be user-friendly for non-bioinformaticians, producing clear, interpretable taxonomic tables.
• Hybrid Virus Detection: Successfully detects both RNA viruses and single-stranded DNA viruses (e.g., Densoviruses) in enriched libraries.

4. Results

• Detection Accuracy: ViroSeek achieved 100% sensitivity in detecting all expected viral species across all five test samples (including low-abundance targets in MixC and the DNA virus in AltMix).
• Contaminant Removal: Effectively removed bacterial and host sequences (e.g., Helicobacter in simulated data and mosquito host RNA in experimental data).
• Performance Comparison:
  • Speed: ViroSeek was significantly faster, requiring ~185 CPU hours for the test set, compared to 3,891 hours for Taxprofiler and 453 hours for MetaDenovo (approx. 4x and 20x faster, respectively).
  • Memory: Memory usage was comparable to competitors (~42 GB max).
  • Sensitivity vs. Specificity: While Taxprofiler detected all expected viruses, it produced a high number of false positives (detecting non-target families) and lower relative abundance for target viruses. MetaDenovo failed to detect several expected arboviruses (e.g., Wesselsbron, West Nile) in diluted samples. VirusTaxo struggled with species-level resolution (>90% of reads unassigned at species level).
• Limitations Observed:
  • Taxonomic Misassignment: Some misclassifications occurred due to database inconsistencies (e.g., AalDV2 assigned to AgDV species; Wesselsbron confused with Sepik virus). The authors note these are database limitations rather than pipeline failures.
  • Cross-Contamination: A single contig for Usutu virus was detected in a sample where it was not introduced, highlighting the need for rigorous wet-lab controls.

5. Significance

ViroSeek fills a critical gap in the virome analysis landscape by providing a robust, reproducible, and accessible tool specifically tailored for second-generation sequencing of target-enriched libraries.

• Public Health Impact: It enables rapid and reliable monitoring of arbovirus emergence, which is vital for disease surveillance in the context of climate change.
• Scientific Utility: By generating assembled contigs and accurate abundance tables, it supports not only detection but also evolutionary studies and phylogenetic tracking.
• Community Adoption: Its open-source nature and simplified workflow lower the barrier to entry for researchers in epidemiology and ecology who may lack advanced bioinformatics expertise, facilitating broader adoption of NGS-based viral surveillance.

The pipeline is available at GitHub (MargauxLefebvre/ViroSeek) and the data is deposited in the European Nucleotide Archive (PRJEB93907).