Life Sciences and Food: OBO, GO, SNOMED CT, FoodOn and AGROVOC

Biology is the only domain that made open ontology work at industrial scale, and it did so by imposing rules on itself that no other sector has been willing to accept. Mandatory open licensing. A single identifier space with permanent URLs. Textual definitions for every term. Versioned releases with an automated build. Deprecation instead of deletion. The result is an ecosystem where a gene annotation made in one laboratory in 2004 still resolves, still means the same thing, and still joins to data produced last week. If you want to know what mature ontology governance looks like, stop reading vendor blogs and read the OBO Foundry principles.

The OBO Foundry: principles as the product

The Open Biological and Biomedical Ontology Foundry is not an ontology. It is a set of principles and a registry of ontologies that agree to follow them. The published principle set currently runs to sixteen numbered principles. The first is openness: the ontology must be openly available to be used by all without constraint, beyond requirements of attribution and non-alteration. The rest are the operational discipline that makes the openness usable: a common formal syntax, a unique OBO Foundry permanent URL identifier space, documented versioning with marked releases, a clearly specified scope, textual definitions for the majority of classes, reuse of existing relations rather than invention, documentation, evidence of a plurality of independent users, collaborative development, a named locus of authority, naming conventions, notification of major changes, active maintenance, term stability, and responsiveness to the community.

Read that list again as a project charter for any domain. Most enterprise ontologies fail on principle nine alone, plurality of users, because they were built for one project and never reused. The Foundry also ships the tooling that makes compliance affordable: the ROBOT command line tool for ontology build operations, the Ontology Development Kit for repeatable release pipelines, and a public dashboard that scores each member ontology against the principles. Governance is automated, which is the only way governance survives contact with deadlines.

Gene Ontology: scale with evidence

The Gene Ontology describes gene product biology across three orthogonal branches: molecular function, biological process and cellular component. Its distinguishing feature is not the ontology but the annotation corpus attached to it. The Gene Ontology knowledgebase report published in Nucleic Acids Research for 2026 records that, as of July 2025, the knowledgebase held 1,135,364 experimentally supported annotations drawn from over 185,000 peer-reviewed publications, with 11,036 of the 11,952 protein families that have experimental data now annotated. Every annotation carries an evidence code and a reference, which means a downstream user can filter by how the claim was established.

The GO Consortium maintains it, releases are frequent and versioned, and the content is openly licensed under CC BY. The deployment is so ubiquitous that biologists forget it is an ontology: GO enrichment analysis, in which a list of differentially expressed genes is tested against GO term annotations, is a standard method reported in tens of thousands of papers, and GO annotations are distributed inside UniProt and every major model organism database. If somebody asks whether ontologies deliver return on investment, this is the strongest single answer in existence. Details of the current release are at the Gene Ontology resource.

SNOMED CT: the licensing reality

SNOMED CT is the world’s most comprehensive clinical terminology, with more than 350,000 active concepts and over a million English descriptions, organised as a description logic terminology that supports post-coordination, meaning you can compose an expression such as a fracture of a specified bone with a specified laterality rather than waiting for a precoordinated code. It is maintained by SNOMED International, which has grown to 53 member countries and has issued affiliate licences to more than 50,000 individuals and organisations.

The licensing is where teams get caught. Use is free in member territories, in low income countries, and for qualifying research projects anywhere. Deploying in a non-member country requires an annual affiliate licence application. That single fact shapes architecture decisions: a multinational product cannot assume it may ship SNOMED content everywhere, so the terminology usually sits behind a service with territory-aware entitlement rather than being baked into a distributed model. This is the sharpest contrast in this lesson. The OBO ecosystem made openness a condition of membership; the clinical world built a members’ club, and the tooling ecosystems reflect that difference precisely.

The deployment record is unambiguous. NHS England mandated SNOMED CT for clinical documentation in primary care systems from April 2018, with a phased rollout to GP practices, and extended the requirement to secondary care, acute care, mental health, community services, dentistry and optometry before 1 April 2020. Every GP consultation recorded in England since then is coded in a description logic terminology, which is a larger production deployment of formal semantics than anything in finance or defence.

FoodOn and AGROVOC: the food and agriculture pair

FoodOn is the consortium-driven food ontology built to interoperate with the OBO Foundry library. It offers over 24,000 terms in its food product hierarchy and description facets, covering applied cooking treatments, preservation methods and packaging, plus roughly 10,000 further terms imported from other OBO ontologies. It is licensed CC BY 4.0. Its purpose is to make food descriptions computable across food safety, traceability, nutrition and agricultural research, where the same product may be called a dozen different things in a dozen different registries.

AGROVOC is the other half of that world and a different kind of artefact: a multilingual thesaurus rather than an axiomatised ontology. Maintained by the Food and Agriculture Organization of the United Nations, the July 2025 release contains 41,447 concepts and 1,219,844 terms in up to 42 languages. It has been published as SKOS since 2009 and as linked open data since 2018, with core and linked open data distributions in RDF plus a public SPARQL endpoint at the AGROVOC browser. Its long-standing deployment is subject indexing for FAO’s bibliographic and information systems, and it functions as the concept hub that national agricultural vocabularies align to.

The pairing teaches a modelling lesson worth carrying into any domain. AGROVOC gives you breadth, multilinguality and stable concept identifiers, but SKOS broader and narrower relations are deliberately weak: they will not tell a reasoner that a pasteurised product underwent a heat treatment process. FoodOn gives you the axiomatic structure to say that, in a single language, over a narrower scope. Serious food supply chain work uses both, joined by mappings, which brings us straight back to the crosswalk discipline from the previous module.

What every other domain should copy

Five practices transfer directly. Make an open licence a condition of inclusion, because a registry with mixed licensing cannot be redistributed and therefore cannot be tooled. Own an identifier space with permanent URLs and never reuse an identifier, so that a fifteen-year-old annotation still resolves. Automate the release: a build that regenerates, reasons over and validates the ontology on every change turns governance from a committee activity into a pull request. Require textual definitions, since a class with a label and no definition is an invitation for two teams to use it differently. Deprecate rather than delete, keeping obsolete terms with a replacement pointer so downstream data degrades gracefully instead of breaking.

None of that is biology-specific. It is release engineering applied to meaning, and it is why one field has a genuinely reusable ontology ecosystem while others have a filing cabinet of PDFs. If you want the formal underpinnings behind these artefacts, from description logic profiles to SKOS semantics, our ontology foundations course covers the standards these communities apply daily.