RgenomicsETL/duckvep-models

DuckVEP transcript models

Transcript models for DuckVEP, a DuckDB extension that annotates variant consequences the way Ensembl VEP does. Each file is a DuckDB database built from an Ensembl core database and the matching primary-assembly FASTA. It holds the regions, transcripts with their exons and coding and flanking sequence, and, where present, the regulatory and motif features that duckvep_model_load reads. The model_receipt table is the model's receipt from duckvep_model_receipt_sql: the source release, the transcript filter, the source and reference hashes, counts, and model_sha256, the model identity that external relations join on.

File	Assembly	Ensembl release	Transcripts	Regions	Regulatory features	Size
ensembl-116/homo_sapiens_116_GRCh38.duckdb	GRCh38	116	644,427	194	1,383,580	1.7G
ensembl-116/homo_sapiens_116_GRCh37.duckdb	GRCh37	116	195,379	84	0	330M

SHA256SUMS lists the checksum of every file.

Use

Download a model, then load it once per DuckDB session:

hf download RgenomicsETL/duckvep-models ensembl-116/homo_sapiens_116_GRCh38.duckdb --repo-type dataset --local-dir models

LOAD 'duckvep.duckdb_extension';
ATTACH 'models/ensembl-116/homo_sapiens_116_GRCh38.duckdb' AS src (READ_ONLY);
SELECT loaded FROM duckvep_model_load('grch38',
 'SELECT seq_region::UINTEGER AS seq_region, sequence_length, seq_region_name FROM src.model_regions ORDER BY seq_region',
 'SELECT transcript_index, seq_region, transcript_start, transcript_end, strand, gene_index, transcript_flags, cds_start, cds_end, cds_sequence, codon_table, pre_cds_sequence, post_cds_sequence FROM src.model_transcripts',
 'SELECT transcript_index, e.exon_start, e.exon_end, e.exon_cdna_start, e.exon_cdna_end, e.phase, e.end_phase FROM src.model_transcripts, unnest(exons) u(e)',
 mature_mirna_query := 'SELECT transcript_index, x.mature_mirna_start, x.mature_mirna_end FROM src.model_transcripts, unnest(mature_mirna_regions) u(x)',
 peptide_edit_query := 'SELECT transcript_index, x.protein_position, x.alternate_amino_acid, x.edit_code FROM src.model_transcripts, unnest(peptide_edits) u(x) ORDER BY 1, 2',
 transcript_coverage_complete := TRUE, interval_feature_query := 'SELECT regulation_feature_index, seq_region, feature_start, feature_end, feature_kind FROM src.duckvep_regulation_features ORDER BY seq_region, feature_start, regulation_feature_index');

duckvep_model_save writes the loaded model to a snapshot that duckvep_model_restore maps back in under a second; snapshots belong to the extension build that wrote them, so they are not published here. The function reference describes loading and annotation.

MANE on GRCh37

MANE is defined only on GRCh38. ensembl-116/mane-grch37/ holds the DuckVEP data release MANE v1.5 mapped to GRCh37, validated against ensembl-116/homo_sapiens_116_GRCh37.duckdb. Its rows join to annotation rows on model_sha256 and transcript_index, and that model's model_receipt holds the matching model_sha256. Label joined rows "MANE mapped to GRCh37", never native MANE. Rows with mapping_status cds_exact_utr_differs speak for the coding region only. The caller recipe shows the join.

Terms

The data come from Ensembl, which makes them available without restriction; see the Ensembl disclaimer. When you use them, cite Ensembl and DuckVEP.