Dataset card: layout, measured counts, format gotchas, mixed licensing

README.md

@@ -0,0 +1,220 @@
+---
+pretty_name: WormBase WS298 — ontologies, associations, and AceDB dumps
+license: cc-by-4.0
+language:
+  - en
+tags:
+  - biology
+  - genomics
+  - c-elegans
+  - caenorhabditis-elegans
+  - wormbase
+  - ontology
+  - gene-ontology
+  - obo
+  - gaf
+size_categories:
+  - 1M<n<10M
+---
+
+# WormBase WS298
+
+A byte-faithful mirror of the parts of WormBase release **WS298** the Liu lab depends on: the five
+ontologies, the seven association files, and two classes of the AceDB dump. Files are exactly as
+WormBase served them — still gzipped, original filenames, nothing decompressed, reorganised or
+re-encoded.
+
+WS298 is WormBase's **final production release**. Ongoing curation moved to the
+[Alliance of Genome Resources](https://www.alliancegenome.org/), so this is a terminal snapshot
+rather than one that will be superseded — which is the reason to pin it rather than track a moving
+`current-production-release/` alias.
+
+## Layout
+
+```text
+WS298/
+├── ontology/     five ontologies + seven association files (12 files, 22 MB)
+└── acedb/        Paper and LongText from the AceDB dump (2 files, 120 MB)
+```
+
+The `WS298/ontology/` path is not arbitrary. It is exactly the cache layout
+[`eutely.wormbase.Release`](https://github.com/liuhlab/eutely) expects, so a snapshot of this repo
+can be read directly as a staged release with no copying or renaming — see [Usage](#usage).
+
+## Ontologies
+
+`anatomy`, `development` and `phenotype` are worm-specific and authored by WormBase. `disease` and
+`gene` are the full Disease and Gene Ontologies, which WormBase mirrors rather than authors.
+
+| File | Terms | IDs | Size |
+| --- | --- | --- | --- |
+| `anatomy_ontology.WS298.obo.gz` | 7,192 | `WBbt:` | 160 KB |
+| `development_ontology.WS298.obo.gz` | 777 | `WBls:` | 29 KB |
+| `phenotype_ontology.WS298.obo.gz` | 2,712 | `WBPhenotype:` | 176 KB |
+| `disease_ontology.WS298.obo.gz` | 14,430 | `DOID:` | 1.3 MB |
+| `gene_ontology.WS298.obo.gz` | 48,165 | `GO:` | 4.5 MB |
+
+All five declare `format-version: 1.2`.
+
+## Associations
+
+An association is one statement that some entity stands in a named relation to one ontology term,
+backed by evidence. Six sources in four wire formats, 1,323,659 statements between them.
+
+| File | Source | Format | Rows | Terms | Size |
+| --- | --- | --- | --- | --- | --- |
+| `anatomy_association.WS298.wb.gz` | `anatomy` | GAF 2.0 | 461,223 | `WBbt:` | 3.7 MB |
+| `development_association.WS298.wb.gz` | `development` | GAF 2.0 | 74,571 | `WBls:` | 382 KB |
+| `phenotype_association.WS298.wb.gz` | `phenotype` | GAF 2.0 | 439,924 | `WBPhenotype:` | 3.1 MB |
+| `gene_association.WS298.wb.gz` | `gene` | GAF 2.2 | 312,203 | `GO:` | 4.4 MB |
+| `disease_association.WS298.daf.txt.gz` | `disease` | DAF 1.0 | 2,364 | `DOID:` | 37 KB |
+| `disease_association.by_orthology.WS298.tsv.txt.gz` | `disease_by_orthology` | TSV | 33,036 | `DOID:` | 222 KB |
+| `gene_association_nonnoctua.WS298.wb.gz` | — | GAF 2.2 | 294,407 | `GO:` | 4.1 MB |
+
+Notes on the columns that differ between sources: in `anatomy`, column 4 is an expression
+*certainty*, not a relation; in `gene` it is a real GAF 2.2 relation (`enables`, `involved_in`, …).
+`disease` is curated from worm experiments and its subject is often an allele or strain rather than
+a gene, while `disease_by_orthology` is inferred from a human ortholog and so carries neither
+evidence code nor reference. The two disease files answer the same question with incomparable
+evidence, which is why they stay separate rather than being concatenated.
+
+`gene_association_nonnoctua` is **not** a seventh source: every one of its rows appears verbatim in
+`gene_association`. It is the GO set with the Noctua/GO-CAM rows removed, included here only so the
+mirror is complete.
+
+**Rows are not records.** A GAF row naming two comma-separated terms is two statements. Parsing the
+seven files yields slightly more records than they have rows — 461,422 records from `anatomy`'s
+461,223 rows, 2,503 from `disease`'s 2,364. The table above counts rows, because that is what is in
+the file.
+
+## AceDB dump
+
+Two classes from the WS298 AceDB dump, kept under their published names.
+
+| File | Contents | Records | Size (gz) | Size (raw) |
+| --- | --- | --- | --- | --- |
+| `Paper.xml.gz` | Bibliography: title, journal, authors, abstract pointer | 64,863 `<Paper>` | 23 MB | 302 MB |
+| `LongText.xml.gz` | Free-text bodies — paper abstracts, EMBL records | 671,404 stanzas | 97 MB | 973 MB |
+
+⚠️ **`LongText.xml.gz` is not XML.** Despite the `.xml.gz` name it is AceDB `.ace` format —
+671,404 `LongText : "<id>"` stanzas of free text, and zero XML tags in the entire file. Feeding it
+to an XML parser fails immediately. `Paper.xml.gz` *is* genuine XML. The misnomer is upstream's;
+the name is preserved here so the file still matches anyone else's copy of the same dump, but do
+not select a parser by suffix.
+
+The two connect through the abstract pointer: a `<Paper>` carries
+`<Abstract><LongText>WBPaper00000003</LongText></Abstract>`, and the matching
+`LongText : "WBPaper00000003"` stanza holds the abstract text.
+
+Neither file records its release internally. They are placed under `WS298/` because that is the
+release they were dumped from, not because anything in the bytes says so.
+
+## Usage
+
+### With `eutely`
+
+The layout is the one [`eutely.wormbase`](https://github.com/liuhlab/eutely) caches into, so a
+snapshot is readable as-is:
+
+```python
+from pathlib import Path
+
+from huggingface_hub import snapshot_download
+from eutely.wormbase import Filter, Release, load_associations, load_ontology
+
+root = Path(snapshot_download("liuhlab/wormbase", repo_type="dataset"))
+ws298 = Release("WS298", root=root)
+
+anatomy = load_ontology("anatomy", release=ws298)
+anatomy["WBbt:0005672"].name                                     # 'AWC'
+len(anatomy.descendants("WBbt:0003681", ("is_a", "part_of")))    # 215 parts of the pharynx
+
+assoc = load_associations("anatomy", release=ws298)
+assoc.by_term("WBbt:0005672")                                    # genes expressed in AWC
+```
+
+Pass `root=` and nothing downloads from WormBase — which is the point on a compute node with no
+outbound network.
+
+### Without `eutely`
+
+Everything is gzipped text; no special reader is required.
+
+```python
+import gzip
+
+with gzip.open("WS298/ontology/gene_association.WS298.wb.gz", "rt") as fh:
+    rows = [line.rstrip("\n").split("\t") for line in fh if not line.startswith("!")]
+```
+
+The dataset viewer does not render these files, and no `configs` block claims it can: OBO, GAF, DAF
+and `.ace` are not tabular formats the Hub can infer, and the tab-separated ones carry `!` comment
+headers rather than a column row. Download and parse them directly.
+
+## Three things that will bite you
+
+**Most phenotype records are negative.** 318,218 of 439,924 assert that a gene does *not* produce a
+phenotype. "The genes for this phenotype" is the wrong question unless you read column 4 — a
+`NOT` qualifier — or filter on it.
+
+**Most GO records are not *C. elegans*.** Only 134,137 of `gene_association`'s 312,203 rows carry
+`taxon:6239` as their primary taxon; the rest are nine other nematodes. (134,390 rows mention
+`taxon:6239` in *either* taxon field, the extra 253 being interacting-taxon entries — a different
+question with a different answer.)
+
+**Anatomy containment is `part_of`, not `is_a`.** Walking `is_a` alone answers nothing about what
+lies inside the pharynx. Widen the relation set when traversing the anatomy ontology; in the
+development ontology, note that `preceded_by` is a temporal ordering rather than a hierarchy.
+
+## Provenance
+
+Downloaded from `downloads.wormbase.org` on 2026-07-26. The ontology and association files came
+through the `releases/current-production-release/ONTOLOGY/` alias, which is the only path WormBase
+serves — the concrete `releases/WS298/` directory the filenames imply answers 403.
+
+Every file was verified as intact gzip and parsed end to end before upload; the record counts in
+the tables above are measured from these exact bytes, not copied from upstream documentation.
+
+## Licensing
+
+This is a redistribution of third-party data with **mixed** terms, labelled `cc-by-4.0` because
+that is the strictest condition in the bundle. Per component:
+
+| Component | Licence |
+| --- | --- |
+| WormBase-authored data (anatomy, development, phenotype; AceDB dumps) | [CC0 1.0](https://creativecommons.org/publicdomain/zero/1.0/) |
+| Gene Ontology and its annotations (`gene_ontology`, `gene_association*`) | [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) — [GO citation policy](https://geneontology.org/docs/go-citation-policy/) |
+| Human Disease Ontology (`disease_ontology`) | [CC0 1.0](https://disease-ontology.org/about/) |
+
+Attribution is therefore required for the GO-derived files and not for the rest. Complying with
+CC BY 4.0 across the whole set satisfies every component. No file here has been modified, so
+attribution to the original creators is the only obligation.
+
+## Citation
+
+Cite the upstream resources, not this mirror:
+
+```bibtex
+@article{wormbase2024,
+  title   = {WormBase 2024: status and transitioning to Alliance infrastructure},
+  journal = {Genetics},
+  volume  = {227},
+  number  = {1},
+  year    = {2024},
+  doi     = {10.1093/genetics/iyae050}
+}
+
+@article{geneontology2023,
+  title   = {The Gene Ontology knowledgebase in 2023},
+  journal = {Genetics},
+  volume  = {224},
+  number  = {1},
+  year    = {2023},
+  doi     = {10.1093/genetics/iyad031}
+}
+```
+
+## Maintenance
+
+Mirrored by the [Liu lab](https://github.com/liuhlab) for [`eutely`](https://github.com/liuhlab/eutely).
+WS298 is terminal, so this repo is a fixed snapshot: it will not be re-cut for new releases.