A pipeline for viral diversity analysis from assembled contigs

---

Overview

ViralQuest v3 detects and characterizes viral sequences from assembled metagenomics or transcriptomics contigs. The pipeline integrates:

• Diamond BLASTx against a curated viral RefSeq database (filter) and NCBI NR (confirmation)
• HMM profiling with RVDB, Vfam, and EggNOG (viral detection) + Pfam (functional annotation)
• BLASTn — local database or online via NCBI qblast; online mode captures accession and query coordinates
• Taxonomy annotation from NCBI viral taxonomy + ICTV
• Sequence clustering by species
• Salmon quantification with reference or de-novo pathway and bundled multi-kingdom housekeeping genes
• Read coverage profiling (minimap2) — per-base depth track with sense/antisense strand split and base-quality colouring, long-read aware (Illumina / Nanopore / PacBio); coverage discontinuities flag potentially chimeric contigs
• Sequence quality checks — dustmasker low-complexity, self-BLASTn repeat detection (direct / inverted, with dot plot), and jellyfish k-mer repetitiveness on confirmed viral contigs
• Sequence scoring — a deterministic heuristic score that always runs (no LLM/API required), plus optional LLM scoring via Ollama (local) or OpenAI / Anthropic / Google APIs (google-genai)
• Self-contained HTML report with interactive genome map (ORF frames + HMM domains + read-coverage track), cluster analysis, taxonomy tree, BLAST tables, sequence-quality panels, and Salmon plots

Paper: https://link.springer.com/article/10.1186/s12859-026-06391-6

---

Installation

1. Create a conda environment

Create a new conda environment to manage all required packages.

conda create -n viralquest python

After create the new env, use conda activate viralquest to run the next steps.

2. Clone the repository and install the environment

git clone https://github.com/gabrielvpina/viralquest.git
cd viralquest
pip install -e .

pip install -e . installs the Python package and registers the CLI entry points.

3. Install dependencies and databases

viralquest-setup

viralquest-setup performs the full first-time setup in a single command:

1. Verifies that pixi is available; installs it automatically if not.
2. Runs pixi install if the environment is not yet built.
3. Downloads reference databases from Zenodo into data/ (HMM profiles: RVDB, Vfam, EggNOG, Pfam; viral Diamond filter DB).

To update or re-download databases independently:

viralquest-download           # skips files already present
viralquest-download --force   # re-downloads all files

---

External databases (user-provided)

These are optional but required for full pipeline runs.

Database	Purpose	Approx. size
nr.dmnd	Diamond BLASTx NR confirmation	~346 GB
BLAST nt	BLASTn local search	~1 TB

Build NR Diamond database:

wget https://ftp.ncbi.nlm.nih.gov/blast/db/FASTA/nr.gz
gunzip nr.gz
diamond makedb --in nr --db nr.dmnd

BLASTn can also run online (--blastn-online); the NR step is optional.

---

Usage

Minimal run:

viralquest \
  -in   SAMPLE.fasta \
  -out  SAMPLE_output \
  -cpu  8

Full run with NR, local BLASTn, Salmon, and local LLM:

viralquest \
  -in            SAMPLE.fasta \
  -out           SAMPLE_output \
  -nr            /path/to/nr.dmnd \
  -n             /path/to/nt \
  --transcriptome host_transcriptome.fasta \
  --reads         R1.fastq R2.fastq \
  --read-type     sr \
  --hk-genes      housekeeping_ids.txt \
  --model-type    ollama \
  --model-name    qwen3:4b \
  --llm-tokens    low \
  -cpu            8 \
  --live

Arguments

Required

Argument	Description
-in	Input FASTA (assembled contigs)
-out	Output directory (created if absent)

Pipeline tuning

Argument	Description
-cpu	CPU threads for Diamond, HMMsearch, BLASTn, and Salmon (default: 2)
--cap3	Run CAP3 assembly before the pipeline
--min-identity	Minimum % identity for intra-cluster BLASTn alignment (default: 90.0)
--min-coverage	Minimum query coverage (%) for a sequence to qualify as a cluster member (default: 50.0); clusters with fewer than 2 qualifying members are dropped
--force	Export all sequences, ignoring viral confirmation filters

Databases

Argument	Description
-nr / --nr-db	Diamond-format NR database (optional)
--nr-block-size	GB of RAM per Diamond database pass (default: Diamond built-in ~2.0)
--nr-index-chunks	Diamond seed-index chunks; lower = fewer disk passes, higher RAM (default: 4)
--nr-tmpdir	Directory for Diamond temporary files (e.g. /dev/shm for RAM disk)
--db-dir	Flat directory containing all five binary database files; overrides the bundled data/ layout

BLASTn (choose one)

Argument	Description
-n / --blastn-local	Path to a local BLAST nucleotide database
--blastn-online	NCBI e-mail for qblast (no local DB required); captures accession and query coordinates
--blastn-online-db	NCBI database for online BLASTn (default: nt)

Read-based analysis (reads → Salmon + coverage + sequence quality)

Supplying --reads enables three read-based modules: Salmon quantification (short reads only), read-coverage profiling (minimap2, all read types), and sequence-quality checks. --read-type is required whenever --reads is used.

Argument	Description
--reads	FASTQ file(s): one = single-end, two = paired-end
--read-type	Read technology, required with --reads: sr (Illumina), ont (Nanopore), pb (PacBio CLR), hifi (PacBio HiFi). Salmon quantification runs for sr only; with ont/pb/hifi the Salmon step is skipped (its short-read mapping is invalid for long reads) and only minimap2 read coverage is produced.
--transcriptome	Host transcriptome FASTA; enables the Salmon reference pathway
--hk-genes	Text file of reference housekeeping gene IDs (one per line) for normalization; requires --transcriptome
--low-memory	Low-RAM mode for the Salmon index (-k 21, and in de-novo mode drops background contigs < 500 bp). Use when salmon index runs out of memory
--skip-salmon	Skip Salmon quantification but still run read coverage (incl. sense/antisense) and sequence-quality checks
--kmer	k-mer size for the jellyfish repetitiveness score in the sequence-quality module (default: 15)

LLM scoring

Argument	Description
--model-type	Provider: ollama, openai, anthropic, google
--model-name	Model identifier (e.g. qwen3:4b, gpt-4o, gemini-2.5-pro)
--llm-tokens	Prompt mode: high (all hits, full details) or low (best hits only, compact)
--api-key	API key for cloud providers (not required for Ollama)

Output / misc

Argument	Description
--live	Rich Live display: ASCII banner + scrolling log box + step progress bar
-v / --version	Print version and exit
-h / --help	Show formatted help and exit

---

Output

SAMPLE_output/
├── diamond/
│   ├── refseq.tsv               # Diamond BLASTx vs viral RefSeq filter
│   └── nr.tsv                   # Diamond BLASTx vs NR (if --nr-db provided)
├── hmm/
│   ├── RVDB.tsv
│   ├── Vfam.tsv
│   ├── EggNOG.tsv
│   └── Pfam.tsv
├── blastn/
│   └── blastn.tsv               # BLASTn hits (local or online)
├── salmon/                      # Salmon output directory (if --reads, short reads)
│   └── vq_quant/quant.sf
├── coverage/                    # per-sequence read-coverage TSVs (if --reads)
│   └── <seq_id>.tsv             #   bin · position · mean/sense/antisense depth
├── seq_quality/                 # sequence-quality signals (if --reads)
│   ├── <seq_id>.tsv             #   per-feature: low-complexity + self-repeats
│   └── summary.tsv              #   per-sequence scalars (low-cx frac, k-mer score…)
├── SAMPLE_viral_contigs.fasta   # confirmed viral sequences
├── SAMPLE_viralquest.json       # full structured report (JSON)
├── SAMPLE_viralquest.html       # self-contained interactive HTML report
└── viralquest.log               # full pipeline log

The HTML report is fully self-contained (single file, no external dependencies) and includes:

• Statistics dashboard — pipeline summary, BLAST/HMM counts, LLM score distribution
• Cluster analysis — sequence grouping with representative sequences and member alignment stats
• Sequence viewer — per-sequence genome map with lane-packed ORF frames (+1/+2/+3/−1/−2/−3), HMM domain overlays, a two-sided read-coverage track (sense up / antisense down, coloured by base quality), tabbed BLAST hit tables (BLASTn / BLASTx-RefSeq / BLASTx-NR), sequence-quality panels (low-complexity bands + repeat dot plot), LLM analysis text, and FASTA export
• Taxonomy tree — D3 radial tree built from phylum → order → family → genus → sequence
• Salmon quantification — viral expression boxplots grouped by cluster, housekeeping gene bar charts per kingdom, host–viral similarity table (EVE detection)

Screenshots

---

Read-based analysis

When --reads is supplied, three modules run on the confirmed viral sequences in addition to the assembly-based analysis. All three are optional and share the same read input; --read-type selects the sequencing technology.

Read coverage

Confirmed viral sequences are used as a small reference and the reads are aligned with minimap2 (technology-aware preset per --read-type: sr → short read, ont → map-ont, pb → map-pb, hifi → map-hifi). A single samtools mpileup pass yields, per base:

• total depth plus a sense / antisense strand split (forward- vs reverse-mapping reads);
• mean base quality (Phred), used to colour the coverage track (green ≥ Q30, amber Q20–29, red < Q20).

In the HTML viewer this renders as a two-sided track under the ORF lanes (sense grows up, antisense down). A roughly uniform profile indicates a well-assembled contig; an internal coverage discontinuity flags a possible chimeric / mis-assembled junction. Per-sequence depth is also exported to coverage/<seq_id>.tsv. Coverage runs for any read type — it is the read signal used when Salmon is skipped for long reads or via --skip-salmon.

Sequence quality

Structural checks on the confirmed viral contigs, written to seq_quality/:

• dustmasker — low-complexity regions (shown as bands in the viewer);
• self-BLASTn — internal direct and inverted repeats, visualised as a dot plot (slope +1 direct, −1 inverted);
• jellyfish — a k-mer repetitiveness score (k set by --kmer, default 15).

These flag assembly artefacts and repeat structure that can otherwise inflate or distort downstream interpretation.

Salmon quantification

Two pathways are available depending on whether a reference transcriptome is provided.

Reference pathway (--transcriptome + --reads): builds a combined Salmon index from viral sequences + bundled multi-kingdom housekeeping genes + user transcriptome. BLASTn of the transcriptome against viral sequences detects potential endogenous viral elements (EVEs).

De-novo pathway (--reads only): BLASTn of bundled housekeeping genes against the assembled contigs identifies normalization anchors; the full contig set is quantified directly.

Bundled housekeeping genes cover seven kingdoms: mammals, arthropods, plants, fish, fungi, bacteria, and nematodes.

---

Scoring

Every run produces a heuristic score for each exported sequence, and can optionally add an LLM score on top.

Heuristic scoring (always on)

A deterministic, rule-based scorer runs automatically at the end of every pipeline — no LLM, no NR database, and no API key required. It scores exactly the set of sequences the report emits, so a vq_score is always present for each exported sequence even in fully offline runs. It requires no configuration or extra arguments.

LLM scoring (optional)

Runs on the same set of confirmed sequences the report emits: NR-confirmed sequences in the --nr-db pathway, or RefSeq/HMM-confirmed (is_viral) sequences when NR is not used — and all sequences under --force. The Google backend uses the google-genai package (google-genai>=1.0.0).

# Local (Ollama) — minimum recommended model: qwen3:4b
--model-type ollama --model-name qwen3:4b --llm-tokens low

# Cloud API
--model-type google  --model-name gemini-2.5-flash --api-key YOUR_KEY --llm-tokens low
--model-type openai  --model-name gpt-4o           --api-key YOUR_KEY --llm-tokens high
--model-type anthropic --model-name claude-sonnet-4-6 --api-key YOUR_KEY --llm-tokens high

Output per sequence: vq_score (0–100), classification (viral-known / viral-unknown / non-viral), and a free-text analysis paragraph.

Setup guide: wiki/Setup-AI-Summary-resource

---

Multi-sample consolidated report (viralquest-report)

viralquest-report merges several individual ViralQuest runs into a single interactive HTML report, with cross-sample clustering of the confirmed viral sequences. Point it at a directory containing one subdirectory per run (each holding its *_viralquest.json):

viralquest-report \
  -in   results_dir/ \
  -out  consolidated_report/

Arguments

Argument	Description
-in / --input	Directory containing ViralQuest result directories (each with a *_viralquest.json)
-out / --outdir	Output directory for the consolidated report
--min-identity	BLASTn identity floor for cross-sample clusters
--min-coverage	BLASTn coverage floor for cross-sample clusters
--blastn-bin	blastn binary to use (default: blastn)
--no-clusters	Skip cross-sample clustering (Overview + Viewer only)
--version	Print version and exit

Screenshots