ViroSync resource bundle¶

The current core resource version is v1.2.0, supplied as resources_v1_2_0_runtime.tar.gz. It includes marker HMMs, Pfam profiles, gene-function profiles, marker and broad reference-proteome DIAMOND databases, taxonomy labels, and marker annotations used for EVE detection, boundary refinement, classification, and gene labels.

The bundle contains 2,917,745 marker proteins, 44,069,720 reference-proteome proteins, 1,086 marker profiles, 937 Pfam screening profiles, 412 gene-function profiles, and 190,562 taxonomy labels. The marker profiles include 33 structure-derived candidates. Of these, 27 have a capsid-type curated function and give capsid evidence; EVE_CAPSID_NOVEL_0029 is the only putative MCP.

Curation of this bundle:

Both DIAMOND databases are dereplicated within each genome. Exact repeats are removed first. DIAMOND 2.2.1 cluster --approx-id 90 --member-cover 80 then groups the remaining proteins of each genome, and the longest member of each cluster is kept. No proteins are merged across genomes. Dereplication leaves 2,917,745 marker proteins and 44,070,936 proteome proteins.
1,258 proteome entries that copy prokaryotic contaminants removed from eukaryotic references in GVClass resource v2.1.0 (Zenodo) are removed with their clusters (1,216 representatives). A match needs at least 95% identity and 80% coverage to an EUK-labelled entry of the same assembly. The marker database holds none of them.
The 3,011 eukaryote-labelled reference hits of the 33 structure-derived profiles carry the label EUK-pEVE__<assembly> in both databases, each with its cluster (1,031 marker and 1,892 proteome representatives). Their label is the lineage of the assembly with every named rank suffixed -pEVE, for example EUK-pEVE|Evosea-pEVE|..., as in GVClass. ViroSync counts EUK-pEVE as viral and never as host (see Methods).
Eukaryotic lineages come from NCBI ranks (scripts/harmonize_taxonomy_labels.py), and 245 rows carry EUK-pEVE lineages.
models/gene_function.hmm and models/gene_function_labels.tsv hold 412 Pfam and VOGDB profiles that label core genes.

release-manifests/resources_v1_2_0/ holds the runtime manifest and the build record (build_provenance.json).

Marker references contain full proteins with identifiers linked to reference genomes. The resource build tree stores these sequences in marker/marker.faa and builds marker/marker.dmnd from that file. The runtime archive contains the DIAMOND database; the matching source companion contains the FASTA. Both encode the same marker reference set.

Update your ViroSync checkout before installing v1.2.0. Older checkouts treat EUK-pEVE hits as unknown and give the 27 capsid-type profiles no capsid evidence.

Setup downloads about 5.3 GB and installs about 11.4 GB. The archive and installed data can use about 17 GB at the same time. Allow extra space for the Pixi environment, results, filesystem overhead, optional resources, and retained older versions.

Install the core resources:

pixi run setup-virosync-resources

To use another location, set the database root before setup and keep it set for later runs:

export VIROSYNC_DB_ROOT=/data/virosync-db
pixi run setup-virosync-resources

Verify the complete installed bundle:

pixi run virosync orchestrate resources verify \
  --config config/orchestration.yaml \
  --full

--full hashes each resource file and checks the DIAMOND databases. Run it after setup, after a storage error, or after copying the resource tree.

Do not run setup while ViroSync is using the databases.

Gene-function profiles (bundle v1.2.0 and later)¶

A runtime bundle of v1.2.0 or later carries two more files next to combined.hmm:

File	Role in the manifest	Content
`models/gene_function.hmm`	`gene_function_models`	HMMER3 text profiles from Pfam and VOGDB
`models/gene_function_labels.tsv`	`gene_function_labels`	One row per profile: `profile`, `source`, `label`, `module`, `max_evalue`, `description`

The manifest counts both files (gene_function_models and gene_function_labels). The counts must be equal, and the profile names in the two files must match. scripts/build_resource_bundle.py parses both files before it writes a manifest. It requires at least one profile, a valid HMMER3 library, a labels table that passes the runtime reader, and the same profile names once each in both files. Installation checks hashes and counts only. Bundles before v1.2.0 keep their file lists. A run on such a bundle has no gene-function files, so it logs one INFO line and writes no gene_functions.tsv. The output reference describes how a run uses the profiles.

The profiles label genes. Labels never change detection or boundaries, and a label that the hallmark catalogue maps to a panel adds compatible context to the score. The label vocabulary follows Fig. 4 of Bellas and Sommaruga (2026, bioRxiv, doi:10.64898/2026.06.19.733378). Each label belongs to one module:

Module	Labels
capsid	MCP, mCP, ATPase, Protease, Portal, Terminase, Triplex
replication	PolB, pPolB, PolA, SF1, SF3, AEP
integration	YR, RVE, Int
mriya	VLTF2, VLTF3, HUH
other	MTase, Tet, SLATT, Lipase
hallmark	RNAPL, TFIIB

A profile enters the set in one of two ways:

Pfam: the model NAME is one of the Pfam families of the vocabulary. A hit counts at an i-Evalue of 1e-5 or lower.
VOGDB: the consensus description of the group, without its REFSEQ prefix, names a gene of the vocabulary, in any virus lineage. A hit counts at a full-sequence E-value of 1e-9 or lower. The label rules exclude look-alikes such as the PCNA "DNA polymerase sliding clamp", the lambda "DNA-packaging protein FI" and RecT recombinases.

Rebuild the two files from a Pfam-A and a VOGDB release:

pixi run python scripts/build_gene_function_profiles.py \
  --pfam Pfam-A.hmm --pfam-release 38.2 \
  --vog vogdb.hmm --vog-annotations vog.annotations.tsv \
  --vog-release <release> --output-dir <dir>

The script streams both HMM files, so it does not load the 4 GB VOGDB file into memory. Copy the two outputs into models/ of the resource tree before scripts/build_resource_bundle.py --split --version v1.2.0.

Credits¶

VOGDB (Virus Orthologous Groups), CC BY 4.0: Trgovec-Greif et al. 2024, Viruses 16:1191, doi:10.3390/v16081191. The profiles in gene_function.hmm keep their VOGDB identifiers.
Pfam, CC0: the Pfam profiles in gene_function.hmm keep their Pfam names.

The bundle README repeats both credits.