RVDB database, protein version

All versions

Full = unclustered protein sequences · Unique = clustered at 100% identity · Profiles = HMM profiles · Annotations = SQLite + plain text files

Annotations

The annotation database assigns functional and taxonomic context to every HMM profile family. It is available both as a SQLite database and as tab-delimited text files.

SQLite schema

Example queries

1. Top keywords for one family

-- List the 10 most frequent keywords for family ID 5
SELECT k.str, fk.freq
FROM fam_kw fk
JOIN keyword k ON fk.kwId = k.id
WHERE fk.famId = 5
ORDER BY fk.freq DESC
LIMIT 10;

2. Largest families

-- List the 10 largest HMM families by number of sequences
SELECT id, size, nbseq, LCAtaxid
FROM family
ORDER BY nbseq DESC
LIMIT 10;

3. Taxonomic composition of a family

-- List taxa associated with family ID 5, ordered by frequency
SELECT taxid, freq
FROM fam_tax
WHERE famID = 5
ORDER BY freq DESC;

4. Families associated with a keyword

-- Count families and total occurrences for the keyword "Influenza"
SELECT COUNT(DISTINCT fk.famId) AS family_count,
       SUM(fk.freq) AS total_occurrences
FROM fam_kw fk
JOIN keyword k ON fk.kwId = k.id
WHERE k.str = 'Influenza';

Plain-text alternative: if you prefer not to use SQLite, the same annotations are provided as tab-delimited text files in U-RVDBvX.X-prot-hmm-txt.tar.xz.

Method

RVDB-prot is produced by a Snakemake pipeline that converts the nucleotide RVDB release into protein sequences, clusters them, builds HMM profiles and annotates every family. The source code is available on Institut Pasteur GitLab.

From nucleotide RVDB to protein flat file

For every entry of the nucleotide RVDB release, the corresponding protein coding sequences are extracted from GenBank/RefSeq records. Each FASTA header keeps a trace of both the protein and the original nucleotide record:

>acc|<p_bank>|<p_acc>|<n_bank>|<n_acc>|<descr[sp]>

where p_bank / p_acc identify the protein source, n_bank / n_acc the original nucleotide source, and descr[sp] the protein description and species.

Clustering and HMM profile generation

The workflow was inspired by vFAM (Skewes-Cox et al.) and re-implemented as a Snakemake pipeline. Protein families are built and converted into HMM profiles through the following steps:

1. Duplicate removal with CD-HIT at 100% identity.
2. All-against-all comparison with BLAST+.
3. Family definition with SiLiX from the similarity network.
4. Filtering of clusters containing at least 4 sequences.
5. Multiple sequence alignment with MAFFT in auto mode.
6. HMM profile building with HMMER hmmbuild using default parameters.

Current build tools: CD-HIT 4.8.1, BLAST+ 2.10.0, SiLiX 1.2.11, MAFFT 7.526, HMMER 3.3.1 and Python 3.11.11.

Annotation of HMM profiles

Each HMM profile family is annotated by combining three sources of information:

• PFAM domains matched with HMMER using --cut_ga, providing functional keywords.
• Sequence descriptions used as a fallback when no PFAM hit is available.
• Taxonomic lineage summarized as a Last Common Ancestor (LCA) from NCBI TaxIDs.

See the Annotations section for the SQLite schema and example queries.