Repository navigation
Releases
GDC viruses source records (200 GenBank and PaVE documents)
Every document a gdc-viruses build requests: 227 verbatim GenBank (gbwithparts text) and PaVE (JSON) documents, one gzipped file each, named
ORFdata 0.42.6
Accept PyEnsembl's canonical ensembl provider name in Proteome.from_ensembl, alongside the display spelling Ensembl.
The supplied context is preserved. Exact assembly/release, source-checksum, sequence-identity and complete-inventory checks remain enforced. This enables the common Vaxrank host builder to share a matching Ensembl annotation with variant inputs. Existing data artifacts are unchanged.
Reviewed in #144; closes #143. Validation: 1,157 local tests passed with two optional skips, all Python 3.11–3.14 CI jobs and extracted-source checks passed, plus lint, strict docs and normal-wheel acceptance. The full 245,535-protein human reference also passed the mixed CTA/variant scope check in the consumer.
ORFdata 0.42.5
Build pinned Ensembl host proteomes with the same source provenance contract as RefSeq, without annotation indexes.
- Add
Proteome.from_ensembl()for exact GTF, cDNA, ncRNA and peptide source bytes. - Validate versioned identities, complete protein inventory, assembly accession and parsed-byte checksums.
- Preserve native internal-stop proteins and explicit untranslated RNA coverage gaps.
- Retain original transcript envelopes without claiming reconstructed CDS coordinates or empirical translation.
Reviewed in #142; closes #141. The full Ensembl 115 reference contains 245,535 proteins and produced a 44.7 MB snapshot in about 64 seconds using existing cached inputs. No annotation indexes or duplicate source downloads were created.
Validation: 1,155 local tests passed (two optional skips), 73 extracted-source checks, lint, strict documentation, distribution checks, normal installed-wheel acceptance, and independent all-protein sequence/identity audit. Public distribution hashes and public-wheel receipts are attached after verification.
ORFdata 0.42.4: native-validated mouse RibORF
Mouse RibORF previously applied native RNA offsets to newer Ensembl transcripts and checked length/start/stop before attaching the original Ribo-seq evidence. Those checks could miss an internal sequence change. Generation 3 reconstructs each ORF against pinned GENCODE M25 RNA and GTF first, then requires exact full-RNA, CDS and protein preservation for a target placement.
The complete study contains 54,128 valid native models: 51,831 have unique Ensembl 107 placements, 118 have multiple exact placements and 2,179 lack an unchanged complete target RNA. Every source BED block and transcript version is checked. Changed/ambiguous cases keep their native model, sequence, evidence and explicit reasons; a surviving stable ID does not establish uniqueness.
Placed records preserve the original in lifted_from and expose a typed ReferenceMapping instead of fabricated chain provenance. Native evidence references remain separate from target annotations through JSON/TSV export, catalogue merging, subsequent chain lifts and extra_orfs(...).unlifted. Superseded mouse generations 1/2 cannot be rebuilt without native validation; generation 3 is the default.
The verified 24.9 MB catalogue is prepared for orfdata data fetch riborf-mouse; rebuilding from the 198 MB pinned source set remains supported. No reference indexes or large vendor databases are needed.
Validation: 1,129 tests passed, two optional skips; 24 final native/mapping regressions passed, including same-length codon changes, compensating indels, version changes, duplicate loci/FASTA IDs, wrong references, replay, TSV and later chain lifts. Lint and strict docs passed. An independent Biopython/base-by-base audit checked all 54,128 native translations and every accepted coordinate path. A normal wheel loaded the complete artifact, reproduced its exact hash on offline replay and passed pip check.
Fixes #87. Code: PR #140. The verified data are published at https://github.com/openvax/orfdata/releases/tag/data-riborf-mouse-3. All final-head CI jobs passed on Python 3.11–3.14, including extracted source archives. PyPI distribution hashes were checked against the reviewed build.
v0.42.3
Adds paired canine mammary-stroma measurements through the common orfdata data commands and QuantitativeDataset API.
- 993,560 native RNA, protein-group and peptide measurements; 28 assay pairs from 14 shared specimens and two missing protein mates.
- A verified 52 MB download, installing 514 MB. Rebuild inputs total about 8 MB.
- Explicit
same_specimenpairing for separate preparations; metadata-derived crosswalk, native missingness, control flags, group ambiguity and incomplete source references remain visible.
Install: pip install orfdata==0.42.3, then orfdata data fetch canine-mammary-stroma.
The study does not establish unique isoform abundance, ncORF translation or MHC presentation. Normal stroma is from cancer-bearing specimens.
PR #139. Data: https://github.com/openvax/orfdata/releases/tag/data-canine-mammary-stroma-2026-10-29.1
Validation: 1,112 local tests passed, two optional skips; full CI and extracted-source tests passed on Python 3.11–3.14. All 993,560 source cells and 6,996 native MS rows were independently reconciled. Ordinary wheel acquisition and offline queries passed. Attached receipts pin the published distributions and data.
ORFdata 0.42.2: prebuilt mouse ncORF catalogue
The existing Chang et al. mouse ncORF catalogue is now a published, checksum-pinned download through the common data API and CLI.
Use orfdata data fetch ncorf-mammals-mouse and orfdata data path ncorf-mammals-mouse. The download and downloads aliases work identically. Explicit source builds remain available.
The 2.22 MB artifact preserves 16,484 ORFs, 32,968 aggregate Ribo-seq summaries, the Ensembl 107/GRCm39 reference, source checksums and one unresolved source row. This packaging release adds no new biological evidence.
Data, provenance and acceptance receipts: https://github.com/openvax/orfdata/releases/tag/data-ncorf-mammals-mouse-1
Validation: 1,103 tests pass, two optional skips; Python 3.11–3.14 CI and extracted-source-distribution checks pass. Lint, strict documentation, public download integrity, complete source-build equality and normal-wheel offline consumer checks pass.
PR: #138
PyPI: https://pypi.org/project/orfdata/0.42.2/
ORFdata 0.42.1
ORFdata 0.42.1 fixes missing carrier orientation in saved TranscriptORF results used by Isovar and Vaxrank.
Every TranscriptORF.to_dict() now includes resolved_transcript.contig and resolved_transcript.strand, including annotated CDSs and TranscriptORF.at placements without catalogue records. The fields describe the actual carrier RNA. Its original source model remains separate, and unknown source-RNA locus values stay null.
Replay accepts older v1 snapshots without these fields, including a frozen snapshot generated by the published 0.42.0 wheel. Supplied locus metadata must match the carrier, with contradictions rejected before reading RNA sequence. Existing sequence, path, reference and fingerprint checks remain enforced.
Review covered both strands, direct and mapped annotated/catalogue ORFs, source RNAs, historical snapshots, contradictory and unknown locus metadata, and input immutability. This release supplies the producer metadata requested in #135; downstream adapters can adopt it without reloading reference sequences.
See PR #137 and the TranscriptORF documentation.
Validation: 1,103 tests passed with two expected skips, 72 focused installed-wheel checks passed with normal dependencies, and pip check, lint, strict docs, wheel/sdist checks and Python 3.11–3.14 push/PR CI passed. The Python 3.14 PR job passed after retrying slow dependency downloads. No reference dataset rebuild was needed.
Post-publication verification: the actual PyPI wheel matches the uploaded artifact checksum and passed all 72 focused checks plus pip check in an isolated environment without Biopython or APSW. Machine-readable acceptance record.
Mouse RibORF: native-validated catalogue, generation 3
Validated mouse RibORF generation 3, built with ORFdata 0.42.4.
All 54,128 native models were reconstructed from pinned GENCODE M25 RNA/GTF and independently audited. 51,831 have a unique exact complete-RNA placement in Ensembl 107; 118 ambiguous and 2,179 changed/missing placements retain their native models and evidence. Ribo-seq evidence stays bound to the original reference. This catalogue does not claim MS or independent-study corroboration.
The 24,875,346-byte gzip catalogue has SHA-256 2d60d7783821f3ab88485197b141c355ca07fbe6e5a9da78cec42262508c5d03. Download with orfdata data fetch riborf-mouse using ORFdata 0.42.4 or later. Rebuilding from pinned sources remains available.
Source: Yang et al., Nature Communications (2024), DOI 10.1038/s41467-024-46240-9, Supplementary Data 3; CC BY 4.0. GENCODE M25 and Ensembl 107 reference sequences retain their original data-use terms and attribution. See the catalogue manifest for exact URLs, sizes and hashes.
Code and review: #140
Mouse ncORF catalogue: Chang et al., version 1
Prebuilt Chang et al. mouse ncORF catalogue, data version 1.
Contains 16,484 ORFs, 16,483 distinct protein sequences, 32,968 aggregate Ribo-seq evidence summaries and one unresolved source row. Reference: Mus musculus GRCm39, Ensembl 107.
The 2,219,659-byte catalog.json.gz is byte-identical to the pinned-source build using released ORFdata 0.42.1. SHA-256: 255f1d089b1eaa88a0bed53006c05ec78350626659461086fa11164947f151f1.
manifest.json records the four source URLs, checksums, access terms and processing. NOTICE.txt includes source attribution and the author dataset's MIT license. acceptance.json records source and local consumer verification.
The data provide aggregate Ribo-seq support, not peptide-MS evidence or per-ORF independent-study counts. Source models and missing evidence are unchanged. The registry integration is in PR #138 for ORFdata 0.42.2.
Citation: Chang et al., Evolutionary remodeling of non-canonical ORF translation in mammals, eLife 2025;14:RP109128. https://doi.org/10.7554/eLife.109128
Paired canine mammary-stroma measurements 2026-10-29.1
Paired canine mammary-stroma RNA/protein measurements from Amini et al. (2021), DOI 10.1016/j.neo.2021.03.001; GEO GSE135183 and PRIDE PXD023023.
Install with ORFdata 0.42.3: orfdata data fetch canine-mammary-stroma.
993,560 native measurements; 18,582 RNA genes × 30 samples; 1,583 protein groups and 5,413 peptide rows × 28 samples. Twenty-eight RNA/protein assay pairs represent 14 shared specimens; two RNA samples retain missing protein mates. Pairing uses a metadata-derived clinical/GEO crosswalk within the authors’ shared cohort, with separate LCM preparations from the same specimen.
Download: 52,014,803 bytes; installed SQLite: 514,154,496 bytes. SHA-256 pins and independent full-source acceptance are attached. The download is a deterministic gzip stream and round-trips to the registered installed hash.
Native groups, controls, zero/NaN missingness and source fields are preserved. Source references remain incomplete; no isoform attribution, gene/protein equivalence, ncORF translation or MHC presentation is asserted. Normal stroma comes from cancer-bearing mammary specimens, not a healthy-animal atlas.
PRIDE quantitative data: CC0. GEO numerical facts and metadata: GEO public-data terms. Factual clinical curation cites Table 1; the CC BY-NC-ND article’s prose and figures are excluded. See the packaged manifest and ORFdata documentation for source pins and terms.