What is KG‑Bioportal?
KG‑Bioportal is a version of the set of ontologies on BioPortal in which ontologies have been transformed to graph nodes and edges in the KGX format. This means it is a collection of entities and relations, with the classes in each ontology serving as the entities and the connections between ontologies becoming relations. Where possible, entities and relations are categorized using Biolink Model, so entries in NCBI Taxonomy are categorized as biolink:OrganismTaxon, and so on.
How is it made?
KG‑Bioportal is made by careful transformation of each ontology from the BioPortal API. Ontology files from BioPortal are transformed to a common format before being converted to nodes and edges. The Summary page reports how many ontologies came through the most recent run, and how large the resulting graphs are.
How are categories assigned?
An OWL file says nothing about Biolink, so every node arrives from the KGX conversion as biolink:NamedThing. KG‑Bioportal then decides a more specific class for as many nodes as the evidence in the graph supports, by four routes, tried in this order:
- Seed terms. A short table of classes whose Biolink meaning is not in doubt: GO's biological_process, MONDO's disease, CHEBI's chemical entity, NCBITaxon's root, and the upper-ontology terms of BFO. A node that is one of these takes its category.
- Inheritance. Every class beneath a seed, by subclass_of or skos:broader, inherits its category. The nearest seed wins, and a domain seed beats an upper-ontology one at the same distance.
- Mappings. An exact_match edge asserts the same referent, so a category crosses it to a node that has none. This is what reaches the UMLS, ICD and SNOMED identifiers an ontology maps to, which have no hierarchy of their own.
- Reviewed roots. Many ontologies hang off roots whose meaning is a fact about that ontology rather than about the term: an ICD chapter, an HGNC locus group, one axis of SNOMED International. Those roots were placed by reading each one and its subclasses in the published graph, with the assistance of an AI agent, one ontology at a time. Each placement is recorded in reviewed_roots.yaml with the reviewer, the date, the graph read, the subclass labels that were the evidence, and whether a maintainer has since confirmed it. Roots that were read and turned down are recorded there too, with the reason. A node that owes its category to a reviewed root is counted apart on its ontology's page, under How they were assigned, so the provenance travels with the number.
A few ontologies are one kind of thing end to end and say so nowhere a machine can read (GNO is glycans; ROR is research organisations). Those take one category as a whole, and the page says so. Nothing else is guessed: a node the evidence says nothing about stays NamedThing. Edges keep the general biolink:Association, because Biolink's association classes cannot be resolved from a category pair alone.
How is it useful?
KG‑Bioportal supports a holistic examination of a broad collection of hierarchical relationships in biology and biomedicine. Because all ontologies are in a common format and data model, they may be merged in a modular fashion and analysed by graph traversal. This enables a growing collection of informative graph machine learning approaches.
What is in the browser?
The graph browser lists two kinds of entry side by side: the BioPortal ontologies KG‑Bioportal has transformed to KGX, which you can download here, and knowledge-graph projects registered in KG‑Registry. Each entry is tagged by source and carries its node and edge counts, version, and download link.