Be aware 1: This put up is a component 3 of a three-part sequence on healthcare, data graphs, and classes for different industries. Half 1, “What Is a Data Graph — and Why It Issues” is offered here. Half 2, “Why Healthcare Leads in Data Graphs” is offered here.
Be aware 2: All photographs by writer
Be aware 3: Whereas doing analysis for this text, I discovered that there are many lists of current sources (ontologies, managed vocabularies, software program) in addition to lists of lists of lists of sources 🤯. So, I constructed an app that runs queries in opposition to Wikidata to get these sources straight. The code is offered here. Let’s use the Semantic Net to energy the Semantic Net :).
Healthcare didn’t grow to be a pacesetter in data graphs by adopting new know-how early. It did so by investing, over centuries, in shared which means. Lengthy earlier than fashionable knowledge platforms or AI, drugs aligned on what exists (ontologies), how entities are named (managed vocabularies), how proof is generated (observations), how knowledge strikes between methods (interoperability requirements), and the way alignment is enforced (by way of regulation, collaboration, and public funding).
This text reveals that healthcare is just not distinctive in needing these foundations, and it’s now not distinctive in constructing them. Different industries are already growing shared ontologies, vocabularies, remark requirements, and change fashions in regulation, finance, local weather science, building, cybersecurity, and authorities. The distinction is just not feasibility, however maturity and coordination.
Within the sections that observe, I stroll by way of the important thing classes different industries can take from healthcare’s expertise, highlighting what healthcare obtained proper, and pointing to concrete examples from different domains the place comparable approaches are already working.
Shared ontologies — agree on what exists
The healthcare trade has tons of ontologies. They’ve ontologies for anatomy (Uberon), genes (Gene Ontology), chemical compounds (ChEBI) and a whole lot of different domains. Repositories corresponding to BioPortal and the OBO Foundry present entry to properly over a thousand biomedical ontologies. Most of those ontologies are area ontologies – they describe the area of healthcare.
Along with these area ontologies, healthcare makes use of cross-domain ontologies like Schema.org and QUDT (Quantities, Units, Dimensions, and Types). They use the Web Ontology Language (OWL), the Shapes Constraint Language (SHACL), and the Simple Knowledge Organization System (SKOS) to construct their ontologies – all requirements from the World Wide Web Consortium (W3C)–extra on this later. There are additionally issues referred to as higher ontologies, that are used to mannequin issues at the next degree than a particular area. Some examples of those are the Basic Formal Ontology (BFO), the Suggested Upper Merged Ontology (SUMO), and gist, a light-weight higher ontology.
Different industries can be taught from healthcare’s historical past of codifying a shared understanding of a website and explicitly agreeing on what exists and the way these issues relate. Whereas healthcare benefited from centuries of empirical science, all industries and organizations take care of entities and guidelines that may be codified. Finance, regulation, provide chains, and even non secular establishments have lengthy relied on formalized constructions to purpose. Listed here are some examples of ontologies being efficiently utilized in different industries:
- The European Legislation Identifier (ELI) Ontology is a powerful instance of a free, publicly funded ontology constructed utilizing W3C requirements. It supplies a shared semantic mannequin for laws throughout EU member states—defining how legal guidelines, amendments, jurisdictions, and authorized relationships are recognized and linked. Moderately than digitizing paperwork alone, it encodes how the authorized system itself works.
- The Environment Ontology (ENVO) is a complementary instance from the scientific neighborhood. ENVO is a community-led, open ontology that represents environments, ecosystems, habitats, and environmental processes. It demonstrates that shared ontologies don’t require centralized authority; they will emerge from distributed skilled consensus and nonetheless grow to be broadly used infrastructure.
- The Financial Industry Business Ontology (FIBO) reveals how finance, like healthcare, advantages from agreeing on core ideas—entities, contracts, and devices—so corporations compete on merchandise reasonably than on definitions.
- EarthPortal is like BioPortal however for Earth sciences, although at a smaller scale. It’s a house for ontologies about Earth sciences, and is basically community-driven, not publicly funded like BioPortal.
- This can be a small subset — for the total record go to this app.
Deal with managed vocabularies as infrastructure, not project-specific
Healthcare superior by treating catalogs of real-world entities as first-class infrastructure. They’ve managed vocabularies for circumstances and procedures (SNOMED CT), illnesses (ICD 11), adversarial results (MedDRA), medication (RxNorm), compounds (CheBI and PubChem), proteins (UniProt), and genes (NCBI Gene). There are even organizations that tie many of those collectively right into a unified data graph just like the Scalable Precision Medicine Open Knowledge Engine (SPOKE), the Monarch Initiative, and Open Targets.
Different industries can do the identical by constructing and curating lists of issues they depend upon (firms, industries, monetary devices, insurance policies, components) and publishing them as open, machine-readable datasets. Listed here are a couple of distinguished examples from different industries:
- The United Nations Bibliographic Information System (UNBIS) Thesaurus is an efficient instance of a free, publicly funded taxonomy that standardizes topics, geographies, and institutional ideas throughout the UN system. It acts as a shared managed vocabulary that permits interoperability throughout companies, studies, and repositories.
- An instance from finance is the Legal Entity Identifier (LEI) system. LEI supplies a world, open identifier for authorized entities taking part in monetary transactions.
- The Worldwide Monetary Reporting Requirements (IFRS) Basis maintains the IFRS Accounting Taxonomy which incorporates parts for tagging monetary statements ready in accordance with IFRS Accounting Requirements.
- AGROVOC is a multilingual managed vocabulary maintained by the Meals and Agriculture Group (FAO) of the United Nations to advertise interoperability of studies and knowledge.
- GeoNames is an open geographic database of over 25 million place names, identifiers, and geographic options. It’s broadly used throughout industries from logistics to information media and is revealed utilizing W3C requirements.
Let empirical remark drive construction
Healthcare advanced by way of remark, experimentation, and replication. Claims about medication should be backed by proof and dogmatists had been (ultimately) overruled by empirical outcomes. In healthcare, the Clinical Data Interchange Standards Consortium (CDISC) standardizes how scientific trial observations—measurements, outcomes, and adversarial occasions—are recorded and evaluated, enabling cumulative, reproducible proof. There are examples of different industries embracing a standardized method to recording observational knowledge:
- The Climate and Forecast Metadata Conventions (CF Conventions) standardize how noticed local weather variables are described throughout sensors and fashions, enabling scientific knowledge to be shared, in contrast, and reused. They’re developed and maintained by way of an open, community-driven course of.
- The Industry Foundation Classes (IFC) from buildingSMART worldwide outline a shared illustration of real-world constructions (buildings, elements, and methods) throughout design, building, and operations. This enables observations about buildings to build up over a construction’s full lifecycle.
Standardize how knowledge is shared, not simply what it means
Healthcare didn’t cease at shared semantics and proof requirements; it additionally standardized interoperability. The Well being Stage Seven Worldwide (HL7) requirements—most notably HL7 FHIR—outline how scientific knowledge corresponding to sufferers, observations, medicines, and encounters are exchanged between methods. Listed here are some examples from different industries:
- The eXtensible Business Reporting Language (XBRL) standardizes how monetary statements and disclosures are reported to regulators and markets. These taxonomies are created by regulators and revealed by way of registries coordinated by XBRL Worldwide
- The National Information Exchange Model (NIEM) is a framework for constructing info schema by aligning on frequent vocabulary and design guidelines throughout domains. This enables details about individuals, occasions, and instances to maneuver between companies or organizations with out shedding which means or authorized integrity.
Use regulation to pressure semantic alignment
Robust regulatory stress compelled healthcare to align on definitions of phrases and requirements for empirical research. The FDA reinforces this alignment by requiring conformity to requirements and managed terminologies, corresponding to CDISC for scientific trial knowledge and MedDRA for adversarial occasion reporting. Different industries, like finance and aviation, are additionally extremely regulated and have standardized methods of reporting and monitoring compliance:
Notably, in healthcare, organizations just like the FDA and WHO actively require using shared vocabularies like MedDRA, ICD, and CDISC in regulatory processes. In finance, whereas regulators just like the SEC and FINRA implement reporting and compliance, there’s not a comparably mature, shared ecosystem of regulatory vocabularies.
Separate pre-competitive semantics from aggressive benefit
Healthcare firms compete on medication, not the definition of medicine. Agreeing on the definition of phrases and finest practices for sharing knowledge doesn’t impede competitors. The Pistoia Alliance exemplifies this method in life sciences by bringing rivals collectively to develop shared semantic requirements and interoperability practices as pre-competitive infrastructure. Listed here are some examples from different industries:
- EDM Council performs a task in finance just like the Pistoia Alliance in life sciences, bringing competing establishments collectively to develop shared knowledge semantics and requirements (together with FIBO) as pre-competitive infrastructure.
- buildingSMART International brings collectively software program distributors, architects, engineers, and building corporations to keep up Business Basis Lessons (IFC). Distributors compete on instruments, however agree on constructing and element phrases and the way in which they’re represented.
- The MITRE Corporation, the R&D group, publishes MITRE ATT&CK, a data graph of adversary ways and methods for choice assist in cybersecurity operations. Whereas safety contractors compete on instruments, they will agree on the language for describing threats and incidents.
Fund shared data as a public good
Public funding has been important for constructing and sustaining healthcare’s ontologies and managed vocabularies, and it’s unlikely that one group would construct all of them by itself. Different industries might construct consortia, foundations, and public-private partnerships to assist an analogous semantic infrastructure. Public funding from the Nationwide Institutes of Well being (NIH) has been important to constructing and sustaining core biomedical ontologies and managed vocabularies. Different industries have additionally benefited from public funding:
Anchor which means in open requirements
Aligning with open requirements ensures that data outlives any single vendor, platform, or know-how. Organizations just like the World Vast Net Consortium (W3C) outline foundational requirements like RDF, OWL, and SHACL. By anchoring semantics in open requirements reasonably than vendor-specific schemas, industries create data that may be reused, built-in, and reasoned over for many years, whilst instruments and architectures evolve.
Writer notice: I function an Advisory Committee member of the World Vast Net Consortium (W3C), an unpaid function held on behalf of my employer, TopQuadrant.
Construct incrementally
Data graphs in healthcare have been the results of a protracted historical past of discovering new issues, documenting the findings, cataloging the cases of courses, and conducting experiments. It’s unlikely that an trade can construct a website data graph top-down. Properly-structured area data can also be not one thing that may be carried out shortly, even with AI.
Conclusion
Lengthy earlier than fashionable knowledge platforms or AI, drugs invested in shared definitions, managed vocabularies, empirical requirements, and interoperable methods of exchanging proof. These decisions allowed data to build up reasonably than fragment.
Different industries don’t want to copy healthcare’s path precisely, however they will undertake a few of its ideas. Agree on what exists. Deal with reference knowledge and vocabularies as shared infrastructure. Let remark and proof drive construction. Use regulation and collaboration to implement alignment. Fund semantics as a public good. Anchor which means in open requirements so it outlives any single vendor or system.
Healthcare didn’t succeed as a result of it adopted AI early. It succeeded as a result of it spent centuries externalizing which means. Data graphs don’t create that settlement—however they lastly make it computable, reusable, and scalable.
Concerning the writer: Steve Hedden is the Head of Product Administration at TopQuadrant, the place he leads the technique for EDG, a platform for data graph and metadata administration. His work focuses on bridging enterprise knowledge governance and AI by way of ontologies, taxonomies, and semantic applied sciences. Steve writes and speaks commonly about data graphs, and the evolving function of semantics in AI methods.
