eduKateSG · Why Science?
Connect many biomedical vocabularies around shared concepts—without pretending their codes, purposes, licences or levels of detail are interchangeable
Choose the closest route, then return to the full index whenever you need the wider evidence chain.
Science learning becomes powerful when students can tell a measured pattern from a model, a statistical decision from a mechanism, and a useful lead from a finished conclusion. The Unified Medical Language System Metathesaurus is produced by the U.S. National Library of Medicine. It is organised by concept, or meaning, and links similar names for the same concept from nearly 200 source vocabularies while preserving the names, identifiers, relationships and views supplied by each source. A Concept Unique Identifier, or CUI, groups synonymous source atoms under the UMLS interpretation; semantic types provide broad categorisation. The current release checked for this article is 2026AA, available on 4 May 2026. A UMLS link is a powerful navigation and integration aid, but it does not make every source code equivalent for billing, clinical care, literature indexing, statistics or research. This guide supports Primary Science, PSLE Science, Secondary Science, O-Level Science and STEM education as a longform evidence-reading route; it is not medical advice, a laboratory protocol, an admission promise or a career guarantee.
Related eduKate reading: MeSH and reproducible biomedical search; SNOMED CT and computable clinical meaning; LOINC and identifiable health measurements; RxNorm and normalised medicine meaning; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 13 October 2026: U.S. National Library of Medicine UMLS Metathesaurus overview, checked 13 October 2026; U.S. National Library of Medicine 2026AA UMLS release notes, released 4 May 2026 and checked 13 October 2026; U.S. National Library of Medicine UMLS overview, checked 13 October 2026; UMLS Reference Manual Metathesaurus chapter, checked 13 October 2026; 2026 Singapore–Cambridge O-Level Biology syllabus; MOE G2/G3 Lower Secondary Science teaching and learning syllabus; 2026 MOE G2 Computing syllabus. Standards, terminologies, classifications and recommendations can change, so preserve dated documentation, exact source data and software versions.
A deeper reading route
Biomedical language is plural by design. A laboratory vocabulary identifies observations, a drug terminology represents products, a disease classification supports statistics, and a thesaurus improves literature retrieval. The UMLS Metathesaurus brings these source vocabularies into a common technical structure so systems can discover shared concepts and related terms without erasing the purpose of each source.
The central unit is the concept. Source strings judged to have the same meaning are assigned to a CUI, while each occurrence from a source remains an atom with its own vocabulary abbreviation, code, term type, language and status. This two-level model supports connection and provenance at once: a common concept for navigation, original atoms for exact source meaning.
A CUI is not a universal replacement code. It expresses the UMLS view that contributing atoms share meaning at a concept level. One source may still distinguish billable subcategories, preferred display terms, inactive codes or context that another lacks. Applications should return to the relevant source vocabulary before making operational decisions.
The Metathesaurus preserves source relationships and can also identify useful relationships among concepts. Hierarchies may differ across sources because they were created for different tasks. An analysis should state whether it followed one named source hierarchy, a UMLS relationship type or a custom graph rule rather than merging every parent link into one imaginary taxonomy.
Semantic types from the UMLS Semantic Network assign broad categories such as disease, substance or procedure to concepts. They help filtering and sense disambiguation, yet they are intentionally general. A semantic type does not carry the detailed clinical definition or workflow constraints of the contributing source.
Release 2026AA became available on 4 May 2026. NLM updated the browser and API for that release and added search filters for Semantic Type and Semantic Group. Reproducible work records the UMLS release, source-vocabulary versions, API or local subset, and any MetamorphoSys configuration used to select content.
Licensing is part of the data model. NLM provides UMLS under a Metathesaurus licence, and some source vocabularies have additional restrictions. A technically available atom is not automatically redistributable. Projects should preserve restriction levels and consult the applicable source terms before publishing extracts.
Mapping is an evidence-producing process. String similarity can generate candidates, CUIs can connect synonyms, and relationships can suggest context. Final mappings may still need source-code granularity, use-case constraints and expert review. Rejected candidates are valuable because they show where labels look alike but meanings differ.
Natural-language processing can use UMLS concepts to normalise mentions, expand queries or support entity linking. Results depend on abbreviation resolution, negation, temporality, subject and document section. Finding a CUI in text does not prove that the patient has the condition or that the author endorses the statement.
For students, UMLS is a grand library catalogue for scientific languages. The cheerful lesson is that connection and difference can coexist: two labels may point to a shared idea, yet the original books still have their own purposes, editions and rules.
MRCONSO.RRF records names and identifiers contributed by sources, while files such as MRREL.RRF represent relationships. A correct database load needs the official field definitions, character encoding, release metadata and licence conditions.
Suppressible, obsolete or non-preferred atoms can remain useful for recognising historical text. Search may include them while display and coding workflows prefer current source terms. The chosen filter policy should be explicit.
Concepts can split or merge as source content and UMLS editing evolve. Historical CUIs therefore need release-aware interpretation. Never assume a CUI observed in one edition has exactly the same atom membership forever.
MetamorphoSys can install and customise the Knowledge Sources. Two teams using the same 2026AA archive can build different subsets if they choose different sources, languages or restrictions. Archive the configuration with the resulting database.
API convenience can hide result ordering, paging and release changes. Store query, filters, endpoint, timestamp and returned identifiers where analyses must be reproducible, and use a small pinned fixture to detect service behaviour changes.
Cross-vocabulary analytics should preserve both the original code and the mapped concept. If a mapping is later revised, researchers can reprocess from source rather than reconstructing information already discarded.
Clinical text and rare concepts can expose sensitive information. Entity linking may increase searchability of records, so access, minimisation and disclosure controls must cover both raw text and derived CUIs.
A terminology graph is not a causal graph. Parent, synonym and related-concept edges support organisation or retrieval, but they do not prove mechanisms, outcomes or treatment effects.
| Stage | Question | Evidence | Responsible output |
|---|---|---|---|
| Preserve | Which source vocabulary statement exists? | Source abbreviation, code, atom, term type and release | A traceable source identity |
| Connect | Which names share a UMLS concept? | CUI, atom membership and editor decisions | A navigable concept bridge |
| Filter | Which meanings and relationships serve the use case? | Semantic type, source, hierarchy and restriction rules | A bounded terminology view |
| Validate | Does the mapping support the intended claim? | Expert review, fixtures, release and provenance | A responsible cross-vocabulary result |
Inside this guide
1–12 · Foundations and design
- 1. Start with the integration question
- 2. Pin UMLS 2026AA
- 3. Treat sources as first-class
- 4. Understand atoms
- 5. Understand CUIs
- 6. Keep labels and identifiers separate
- 7. Read term types
- 8. Read languages explicitly
- 9. Use semantic types broadly
- 10. Use semantic groups for orientation
- 11. Preserve source hierarchies
- 12. Interpret relationship provenance
13–24 · Models and estimation
- 13. Do not equate every shared CUI
- 14. Do not infer synonymy from string equality
- 15. Do not infer causality from graph edges
- 16. Use source restriction levels
- 17. Build a documented subset
- 18. Use the API reproducibly
- 19. Handle inactive content
- 20. Track concept history
- 21. Evaluate mappings by grain
- 22. Retain rejected candidates
- 23. Connect MeSH deliberately
- 24. Connect LOINC and SNOMED deliberately
25–36 · Diagnostics and interpretation
- 25. Connect RxNorm deliberately
- 26. Protect linked clinical text
- 27. Preserve provenance
- 28. Connect to Primary Science
- 29. Connect to Secondary Science
- 30. Connect to mathematics
- 31. Connect to computing
- 32. Connect to pathways
- 33. Did you know?
- 34. Write a reader-ready methods note
- 35. Build a validation ladder
- 36. Finish with a reproducible bundle
Section 1 of 36
1. Start with the integration question
Define whether the task is search expansion, code mapping, text annotation, cohort discovery or terminology research.
Working checkpoint — UMLS Metathesaurus: Write the target use and required source granularity. Keep source atom, source code, CUI, semantic type, relationship provenance, release, mapping method and restriction together.
For a classroom model, group invented synonyms while retaining their source cards.
Section 2 of 36
2. Pin UMLS 2026AA
NLM released the 2026AA Knowledge Sources on 4 May 2026.
Working checkpoint — Concept Unique Identifier: Store UMLS and source-vocabulary releases together. Read every concept beside its contributing vocabularies, term types, languages, status and intended use.
For future study, connect medicine to linguistics and data science while checking pathways separately.
Section 3 of 36
3. Treat sources as first-class
The Metathesaurus is assembled from vocabularies built for different communities and purposes.
Working checkpoint — source vocabulary: Keep source abbreviation and original code on every atom. Preserve original codes and rejected candidates; concept linking is not silent source replacement.
For writing, cite UMLS release, sources, CUI policy, mapping review and licence limits.
Section 4 of 36
4. Understand atoms
An atom is a source-specific occurrence of a term, carrying language, term type, identifier and status information.
Working checkpoint — atom: Inspect atoms before describing a concept. Separate terminology connection from operational equivalence, clinical truth and causal inference.
For students, ask how a catalogue can connect names without making every book the same.
Section 5 of 36
5. Understand CUIs
A Concept Unique Identifier groups atoms that UMLS represents as the same concept meaning.
Working checkpoint — semantic type: Use the CUI as a bridge, not a replacement for source codes. Treat source selection, release and hierarchy rule as scientific inputs: one filter can change the graph.
For parents, ask whether a health-data claim preserves source code and release beside a mapped concept.
Section 6 of 36
6. Keep labels and identifiers separate
A preferred UMLS name supports display, while operational identity may depend on CUI, source code or source atom.
Working checkpoint — biomedical terminology mapping: Store identifier system explicitly. Keep source atom, source code, CUI, semantic type, relationship provenance, release, mapping method and restriction together.
For a classroom model, group invented synonyms while retaining their source cards.
Section 7 of 36
7. Read term types
Source term types indicate roles such as preferred, synonym, entry or code-specific names under source conventions.
Working checkpoint — UMLS Metathesaurus: Use the source documentation for exact meaning. Read every concept beside its contributing vocabularies, term types, languages, status and intended use.
For future study, connect medicine to linguistics and data science while checking pathways separately.
Section 8 of 36
8. Read languages explicitly
The Metathesaurus contains multiple languages, but coverage and synonymy differ across sources.
Working checkpoint — Concept Unique Identifier: Preserve language and translation provenance. Preserve original codes and rejected candidates; concept linking is not silent source replacement.
For writing, cite UMLS release, sources, CUI policy, mapping review and licence limits.
Section 9 of 36
9. Use semantic types broadly
Every UMLS concept receives at least one semantic type from the Semantic Network.
Working checkpoint — source vocabulary: Filter with semantic types without treating them as fine clinical codes. Separate terminology connection from operational equivalence, clinical truth and causal inference.
For students, ask how a catalogue can connect names without making every book the same.
Section 10 of 36
10. Use semantic groups for orientation
Semantic groups collect related semantic types for broader filtering and explanation.
Working checkpoint — atom: Name the grouping version used. Treat source selection, release and hierarchy rule as scientific inputs: one filter can change the graph.
For parents, ask whether a health-data claim preserves source code and release beside a mapped concept.
Section 11 of 36
11. Preserve source hierarchies
Parent and child relationships can express a source vocabulary’s view. Different sources may organise concepts differently.
Working checkpoint — semantic type: Select one documented hierarchy for each analysis. Keep source atom, source code, CUI, semantic type, relationship provenance, release, mapping method and restriction together.
For a classroom model, group invented synonyms while retaining their source cards.
Section 12 of 36
12. Interpret relationship provenance
Relationship files can contain source-provided and UMLS relationships with attributes.
Working checkpoint — biomedical terminology mapping: Inspect source, relationship label and direction. Read every concept beside its contributing vocabularies, term types, languages, status and intended use.
For future study, connect medicine to linguistics and data science while checking pathways separately.
Section 13 of 36
13. Do not equate every shared CUI
Atoms grouped under one concept can still differ in operational use, granularity or validity state.
Working checkpoint — UMLS Metathesaurus: Validate against the destination workflow. Preserve original codes and rejected candidates; concept linking is not silent source replacement.
For writing, cite UMLS release, sources, CUI policy, mapping review and licence limits.
Section 14 of 36
14. Do not infer synonymy from string equality
The same abbreviation can have several biomedical meanings, while true synonyms can look very different.
Working checkpoint — Concept Unique Identifier: Use context, source and concept evidence. Separate terminology connection from operational equivalence, clinical truth and causal inference.
For students, ask how a catalogue can connect names without making every book the same.
Section 15 of 36
15. Do not infer causality from graph edges
Terminology relationships organise meaning; they are not evidence that one condition causes another.
Working checkpoint — source vocabulary: Reserve causal claims for appropriate studies. Treat source selection, release and hierarchy rule as scientific inputs: one filter can change the graph.
For parents, ask whether a health-data claim preserves source code and release beside a mapped concept.
Section 16 of 36
16. Use source restriction levels
UMLS distributes source information under licences that can impose different use and redistribution conditions.
Working checkpoint — atom: Carry restriction metadata into export review. Keep source atom, source code, CUI, semantic type, relationship provenance, release, mapping method and restriction together.
For a classroom model, group invented synonyms while retaining their source cards.
Section 17 of 36
17. Build a documented subset
MetamorphoSys can select sources, languages and options for a local Metathesaurus installation.
Working checkpoint — semantic type: Archive the configuration and output hashes. Read every concept beside its contributing vocabularies, term types, languages, status and intended use.
For future study, connect medicine to linguistics and data science while checking pathways separately.
Section 18 of 36
18. Use the API reproducibly
The UTS browser and API reflect the current release and request parameters.
Working checkpoint — biomedical terminology mapping: Store endpoint, query, filters, paging and timestamp. Preserve original codes and rejected candidates; concept linking is not silent source replacement.
For writing, cite UMLS release, sources, CUI policy, mapping review and licence limits.
Section 19 of 36
19. Handle inactive content
Historical and suppressible atoms can aid recognition but may not be suitable for new coding.
Working checkpoint — UMLS Metathesaurus: Separate search coverage from current-code selection. Separate terminology connection from operational equivalence, clinical truth and causal inference.
For students, ask how a catalogue can connect names without making every book the same.
Section 20 of 36
20. Track concept history
CUI membership and source content can change across releases.
Working checkpoint — Concept Unique Identifier: Compare editions before longitudinal joins. Treat source selection, release and hierarchy rule as scientific inputs: one filter can change the graph.
For parents, ask whether a health-data claim preserves source code and release beside a mapped concept.
Section 21 of 36
21. Evaluate mappings by grain
A broad disease concept, billable code and research phenotype may be related without being interchangeable.
Working checkpoint — source vocabulary: Score exact, broader, narrower and unresolved outcomes separately. Keep source atom, source code, CUI, semantic type, relationship provenance, release, mapping method and restriction together.
For a classroom model, group invented synonyms while retaining their source cards.
Section 22 of 36
22. Retain rejected candidates
Candidate-generation systems can produce plausible but wrong matches.
Working checkpoint — atom: Keep competing candidates and the deciding evidence. Read every concept beside its contributing vocabularies, term types, languages, status and intended use.
For future study, connect medicine to linguistics and data science while checking pathways separately.
Section 23 of 36
23. Connect MeSH deliberately
MeSH supports biomedical literature indexing and search; UMLS can connect its terms with other vocabularies.
Working checkpoint — semantic type: Preserve the MeSH descriptor and year for retrieval. Preserve original codes and rejected candidates; concept linking is not silent source replacement.
For writing, cite UMLS release, sources, CUI policy, mapping review and licence limits.
Section 24 of 36
24. Connect LOINC and SNOMED deliberately
LOINC identifies observations and SNOMED CT represents clinical meaning, while UMLS can expose conceptual links.
Working checkpoint — biomedical terminology mapping: Do not collapse observation identity and result value into one code. Separate terminology connection from operational equivalence, clinical truth and causal inference.
For students, ask how a catalogue can connect names without making every book the same.
Section 25 of 36
25. Connect RxNorm deliberately
RxNorm normalises medicine concepts and also appears as a UMLS source.
Working checkpoint — UMLS Metathesaurus: Use RxNorm release and term type for medication workflows. Treat source selection, release and hierarchy rule as scientific inputs: one filter can change the graph.
For parents, ask whether a health-data claim preserves source code and release beside a mapped concept.
Section 26 of 36
26. Protect linked clinical text
Derived concepts can make sensitive narratives easier to locate and combine.
Working checkpoint — Concept Unique Identifier: Apply purpose-based access and disclosure review. Keep source atom, source code, CUI, semantic type, relationship provenance, release, mapping method and restriction together.
For a classroom model, group invented synonyms while retaining their source cards.
Section 27 of 36
27. Preserve provenance
A reusable mapping links source atom, CUI, release, algorithm, reviewer, restriction and validation.
Working checkpoint — source vocabulary: Make every crosswalk decision reconstructable. Read every concept beside its contributing vocabularies, term types, languages, status and intended use.
For future study, connect medicine to linguistics and data science while checking pathways separately.
Section 28 of 36
28. Connect to Primary Science
A fictional biomedical library-catalogue card helps pupils separate an observation from the standardised name or code used to organise it.
Working checkpoint — atom: Use invented examples and never use personal health information. Preserve original codes and rejected candidates; concept linking is not silent source replacement.
For writing, cite UMLS release, sources, CUI policy, mapping review and licence limits.
Section 29 of 36
29. Connect to Secondary Science
UMLS Metathesaurus supports lessons about classification, evidence, systems, uncertainty and responsible comparison.
Working checkpoint — semantic type: Trace one coded statement back to its observable evidence. Separate terminology connection from operational equivalence, clinical truth and causal inference.
For students, ask how a catalogue can connect names without making every book the same.
Section 30 of 36
30. Connect to mathematics
Sets, hierarchies, identifiers, denominators and versioned mappings show why definitions must be fixed before quantities are compared.
Working checkpoint — biomedical terminology mapping: Work through one small boundary case by hand. Treat source selection, release and hierarchy rule as scientific inputs: one filter can change the graph.
For parents, ask whether a health-data claim preserves source code and release beside a mapped concept.
Section 31 of 36
31. Connect to computing
Relational files, concept graphs, APIs, lexical tools, subset builders and validators show how software connects scientific language without erasing sources.
Working checkpoint — UMLS Metathesaurus: Break one identifier, release, relationship or mapping, then validate and repair it transparently. Keep source atom, source code, CUI, semantic type, relationship provenance, release, mapping method and restriction together.
For a classroom model, group invented synonyms while retaining their source cards.
Section 32 of 36
32. Connect to pathways
UMLS connects medicine, biology, library science, linguistics, ontology engineering and data science and can illuminate STEM education and career pathways.
Working checkpoint — Concept Unique Identifier: Check current official education and qualification routes separately from this conceptual guide. Read every concept beside its contributing vocabularies, term types, languages, status and intended use.
For future study, connect medicine to linguistics and data science while checking pathways separately.
Section 33 of 36
33. Did you know?
The Metathesaurus links similar names by concept while preserving the names and relationships supplied by nearly 200 source vocabularies.
Working checkpoint — source vocabulary: Explain the fact in plain language, then name what it does not prove. Preserve original codes and rejected candidates; concept linking is not silent source replacement.
For writing, cite UMLS release, sources, CUI policy, mapping review and licence limits.
Section 34 of 36
34. Write a reader-ready methods note
State the standard and release, source data, coding or mapping method, validation, privacy controls and supported claim.
Working checkpoint — atom: Keep the note beside the machine-readable record. Separate terminology connection from operational equivalence, clinical truth and causal inference.
For students, ask how a catalogue can connect names without making every book the same.
Section 35 of 36
35. Build a validation ladder
Check identifiers, release, syntax, hierarchy, source fidelity, scientific plausibility, privacy and known fixtures in sequence.
Working checkpoint — semantic type: Stop at the earliest unresolved layer. Treat source selection, release and hierarchy rule as scientific inputs: one filter can change the graph.
For parents, ask whether a health-data claim preserves source code and release beside a mapped concept.
Section 36 of 36
36. Finish with a reproducible bundle
Deliver source receipts, releases, mappings, software versions, tests, hashes and limitations without exposing governed health data.
Working checkpoint — biomedical terminology mapping: Make every published claim traceable. Keep source atom, source code, CUI, semantic type, relationship provenance, release, mapping method and restriction together.
For a classroom model, group invented synonyms while retaining their source cards.
Contents · Previous section · Continue to the Science Learning Hub