Presentation is loading. Please wait.

Presentation is loading. Please wait.

Summary Issues and Suggestions Workshop on The Future of the UMLS Semantic Network NLM, April 8, 2005 Olivier Bodenreider Lister Hill National Center for.

Similar presentations


Presentation on theme: "Summary Issues and Suggestions Workshop on The Future of the UMLS Semantic Network NLM, April 8, 2005 Olivier Bodenreider Lister Hill National Center for."— Presentation transcript:

1 Summary Issues and Suggestions Workshop on The Future of the UMLS Semantic Network NLM, April 8, 2005 Olivier Bodenreider Lister Hill National Center for Biomedical Communications Bethesda, Maryland - USA

2 Issues

3 3 UMLS Semantic Network u Necessary complement to the Metathesaurus l Provides direct categorization to concepts (some of which would be orphans otherwise) u Best used in conjunction with the Metathesaurus u Used for l Natural Language Processing l Information retrieval l Knowledge discovery u Essentially stable

4 4 Semantic types u Purposely limited to a small number of categories u Purposely emphasizes categories of major interest l e.g., Neoplastic Process l No attempt to anything JEPD u No explicit classificatory principles or properties l Textual (not formal) definitions l Introduction points for semantic relationships

5 5 Semantic relations u Single-inheritance hierarchy u Class-class relations u Simply mirrored by inverses u Weakest reading possible: some-some l Sufficient for some applications (e.g., semantic interpretation, reporting and visualization of clinical information) l Too limited for reasoning

6 6 Semantic groups u 15 collections of semantic types u Created for visualization purposes u Purposely non-ontological (not subtrees from the isa hierarchy of STs) u Based on common properties of (sometimes) otherwise heterogeneous semantic types

7 7 Semantic categorization u Generally corresponds to isa (rarely is an instance of) u Convenient for extracting a class l Direct access: no traversal necessary l Bypasses hierarchies in vocabularies: not subject to questionable hierarchical relations

8 8 Semantic type assignment (1) u Essentially manual (default based on source information, reviewed by Metathesaurus editors) u Complex and labor intensive u Multiple ST assignment sometimes required l Structure + role (chemicals) l Systematic polysemy u Guidelines l Usage notes l Prior categorization of similar concepts

9 9 Semantic type assignment (2) u No constraints based on mandatory consistency between SN and Metathesaurus (e.g., ST of the child concept must be identical to or a descendant of ST of the parent concept) u No constraints based on ontological principles (e.g., disjunction between Entity and Event) u No constraints based on structural principles (e.g., allowable hybrid types)

10 10 Systematic polysemy (splitting vs. lumping) u Metathesaurus (RxNorm) distinguishes between l Clinical drug (e.g., Acetaminophen) l Branded drug (e.g., Tylenol) u But does not systematically distinguish between l Prostatic adenoma (the tumor responsible for compressing the urethra) l Prostatic adenoma (the disease of which urinary problems are one manifestation) both contain acetaminophen as their active ingredient

11 11 Finding u Role played by many different types u Necessarily some-some (rare exceptions) u Reified for convenience

12 12 Overall constraints for changes u Finite amount of resources u Driven by usefulness

13 Suggestions

14 14 SN and Metathesaurus u Issues in the SN cannot be dissociated from issues in the Metathesaurus u Inaccurate/inconsistent concept categorization l May be a bigger issue than issues identified in the SN n Relatively frequent n Impair semantic integration and semantic interpretation l Will not be solved solely be addressing issues in the SN

15 15 SN vs. Biomedical ontology u Having a good (high-level) ontology of biomedicine is certainly desirable… u But it will be of little use if it is not linked to Metathesaurus concepts u Some ontological features (e.g., some-all) require a much finer granularity than that of the current semantic types

16 16 Editing vs. Auditing u Auditing must be pursued, but… u Better editing environments are needed l Law: explicit classificatory principles and properties l Order: n Enforce SN/Meta consistency (use SN relations as a reference for Meta relations) n Restrict allowable combinations of STs u Quality assurance starts at the time of editing

17 17 Source transparency vs. Anarchy (1) u All relations asserted by sources are recorded… (source transparency) u But need not be necessarily trusted u Similar to how synonymy is treated l Metathesaurus synonymy does not always follow source synonymy

18 18 Source transparency vs. Anarchy (2) u Similar to how names lacking face validity are treated l Fully specified Metathesaurus names are created l Invalid names are made suppressible u Similarly for relations l Metathesaurus hierarchical relations should ignore some obviously non-hierarchical relations used to create hierarchies in source vocabularies l Suppressibility or Content View Flag (CVF)

19 Agenda

20 20 Semantic types u Rename some types (face validity) u Extract explicit classificatory principles u Rearrange hierarchy as needed (e.g., Alga) u Revisit roles l Place under sortals when unique (e.g., Enzyme) l Create allowable hybrids (e.g., Steroid hormone)

21 21 Semantic relations u Align with Metathesaurus relations (e.g., caused_by / due_to) u Multiple inheritance (?) u Two levels l Coarse class-class, some-some, with mirrored inverses to label the relation (and support semantic interpretation) l Finer non-symmetric class-class, some-all (?) to support reasoning

22 22 ST assignment u Facilitated by improved editing environment u Driven by explicit classificatory principles and properties u Simplified by allowable hybrids u Constrained by coherence with SN relations (requires aligned relations and labeled Metathesaurus relations)


Download ppt "Summary Issues and Suggestions Workshop on The Future of the UMLS Semantic Network NLM, April 8, 2005 Olivier Bodenreider Lister Hill National Center for."

Similar presentations


Ads by Google