Publish BiolinkAmalgam build

README.md

@@ -1,5 +1,27 @@
 ---
 pretty_name: BiolinkAmalgam
+configs:
+  - config_name: default
+    default: true
+    data_files:
+      - split: train
+        path: biolinkamalgam.parquet
+  - config_name: crossrefs_dbsnp_reference
+    data_files:
+      - split: train
+        path: crossrefs/dbsnp_reference.parquet
+  - config_name: crossrefs_gene_association
+    data_files:
+      - split: train
+        path: crossrefs/gene_association.parquet
+  - config_name: crossrefs_replaced_by
+    data_files:
+      - split: train
+        path: crossrefs/replaced_by.parquet
+  - config_name: crossrefs_xref
+    data_files:
+      - split: train
+        path: crossrefs/xref.parquet
 license: other
 license_name: babel-upstream-licenses
 task_categories:
@@ -25,6 +47,14 @@ every GTDB rank becomes a `GTDB:`-prefixed taxon CURIE, and species carrying an
 `ncbi_species_taxid` crossref into their `NCBITaxon:` clique (visible as a `GTDB:`-prefixed
 synonym; at most one GTDB species per NCBI species — many-to-one mappings stay singletons).
 
+ClinVar `variant_summary` contributes `biolink:SequenceVariant` cliques: one `CLINVAR:` row per
+variation and one `DBSNP:rs...` row per referenced rsID. An rsID can denote several alleles, so the
+two are never merged; their links live in the `crossrefs_dbsnp_reference` config.
+
+Additional OBO ontologies not in BABEL (`config/obo_sources.yaml`: MP, ZP, FYPO, WBPhenotype,
+APO, TO, MAXO, HsapDv, MmusDv, ZFS, FBdv, WBls, MOD, TAXRANK, FAO) contribute active own-prefix
+terms with their names, EXACT synonyms and alt_ids; BROAD/NARROW/RELATED synonyms are excluded.
+
 Synonyms are the ORIGINAL strings with junk filtered out (see below); normalization
 (stemming, level keys) is deliberately left to downstream consumers.
 
@@ -38,11 +68,34 @@ One row per surviving clique leader.
 | `PREFERRED_NAME` | string | the clique's preferred name, as written |
 | `CLASS` | string | Biolink category, e.g. `biolink:Gene` |
 | `SYNONYMS` | list<string> | synonyms as originally written, sorted; a synonym appears in at most one row per (class, taxon scope) |
+| `TAXA` | list<string> | the clique's NCBITaxon CURIEs, sorted; empty when taxonless. Taxa come from BABEL, or are inferred from single-species identifier namespaces (MGI, RGD, Ensembl species prefixes) |
+
+## Crossrefs (opt-in conflation)
+
+Typed edges kept OUT of the main table, one config (train split) per predicate, so you choose which
+edge types to conflate. Columns: `subject, predicate, object, source`. Edges carry source CURIEs, not
+post-merge leaders; an obsolete term may appear only here.
+
+ config | file | edge |
+---|---|---|
+ `crossrefs_dbsnp_reference` | `crossrefs/dbsnp_reference.parquet` | ClinVar variation -> dbSNP rsID (one rsID can cover several alleles) |
+ `crossrefs_gene_association` | `crossrefs/gene_association.parquet` | ClinVar variation -> NCBIGene it sits in (location, not identity) |
+ `crossrefs_replaced_by` | `crossrefs/replaced_by.parquet` | obsolete OBO term -> its successor |
+ `crossrefs_xref` | `crossrefs/xref.parquet` | OBO term cross-reference (relevance only per OBO 1.4) |
+
+```python
+dbsnp_reference = load_dataset("SkyeAv/biolink-amalgam", "crossrefs_dbsnp_reference", split="train")
+gene_association = load_dataset("SkyeAv/biolink-amalgam", "crossrefs_gene_association", split="train")
+replaced_by = load_dataset("SkyeAv/biolink-amalgam", "crossrefs_replaced_by", split="train")
+xref = load_dataset("SkyeAv/biolink-amalgam", "crossrefs_xref", split="train")
+```
 
 ## What was filtered
 
 - `biolink:Publication` and `biolink:Agent` cliques (titles and leftover-UMLS organizations).
 - Taxa outside the top-100 experimental taxon allowlist (taxonless and `biolink:OrganismTaxon` cliques kept).
+  Taxonless cliques whose identifiers are single-species (MGI, RGD, ZFIN, FB, WormBase, SGD, HGNC,
+  Ensembl species prefixes; `config/curie_taxa.yaml`) get that taxon first, so they meet the same cutoff.
 - Names that are pure numbers (including decimals like `0.5`), contain `language`,
   `uncharacterized` or `hypothetical`, or equal a medically meaningless stopword
   (`unknown`, `none`, `the`, ...; full list in `config/filters.yaml`).
@@ -52,11 +105,12 @@ One row per surviving clique leader.
 
 | stage | rows | time |
 |---|---|---|
-| ingest     |    140432548 |  3028096 ms |
- merge      |    139080106 |  2822524 ms |
- arbitrate  |    139080106 |  1098051 ms |
+| ingest     |    140432548 |     6701 ms |
+ ingest_extra |      7369667 |   106809 ms |
+ merge      |    141327506 |   926304 ms |
+ arbitrate  |    141327506 |   867696 ms |
 
-Generated 2026-10-24 by [BiolinkAmalgam](https://github.com/SkyeAv/BiolinkAmalgam).
+Generated 2026-10-27 by [BiolinkAmalgam](https://github.com/SkyeAv/BiolinkAmalgam).
 
 ## Usage
 
@@ -76,3 +130,5 @@ BABEL aggregates many upstream biomedical resources with heterogeneous licenses
 upstream terms; check the [BABEL license summary](https://github.com/TranslatorSRI/Babel)
 for your intended use.
 GTDB data are CC-BY-SA-4.0 (share-alike); GTDB-derived taxa carry those terms.
+ClinVar and dbSNP are NCBI public-domain data.
+OBO ontologies carry CC-BY-3.0, CC-BY-4.0 or CC0-1.0 terms per ontology (see config/obo_sources.yaml).

biolinkamalgam.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cbeb6c11a9ba7021e860d187b28c6481645a910251d6c7265c3992ffd2aaf07a
-size 8670218428
+oid sha256:25ebaf28b4541e5f7f2e9993573e78982e373dcb627f4d8365eae24362170785
+size 8897142578

crossrefs/dbsnp_reference.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5663306eba5c67d5796c6f35b874674ffae095057c8022a72a0ea459b227856d
+size 18453851

crossrefs/gene_association.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:999b6722f32c812616d73a9d33da28d0b09393ee1a07b37d807b4392ef9ef23b
+size 10706980

crossrefs/replaced_by.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0ec4e085b4b1619568feebe1e7639e484d9d5b43ee4696f4364ca545e9ad071
+size 3768

crossrefs/xref.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b96b26be6bcd15321a8c0bb648b29d5639647d70fec0e4b5f5f221022dc3faf9
+size 21901