The UMLS Semantic Network

Alexa T. McCrayCenter for Clinical Computing

Beth Israel Deaconess Medical CenterHarvard Medical School

mccray@bidmc.harvard.edu

The Future of the UMLS Semantic Network National Library of Medicine, April 7, 2005

UMLS ProjectUMLS Project

• Begun in 1986

- Well before the advent of the World Wide Web

• Goal

- To provide intelligent access to biomedical resources in multiple, disparate databases• Language of those resources of primary interest

• Methodology

- Consultation with broad medical informatics constituency

- Development of Knowledge Sources

Initial EffortsInitial Efforts

• First versions of knowledge sources available to researchers in early 1990’s- Metathesaurus (1990)

• Interrelate existing vocabularies, thesauri

- Semantic Network (1990)• Assignment of semantic types to Metathesaurus concepts

- Information Sources Map (1991)• Characterization of existing databases, including query

syntax and MeSH indexing

- SPECIALIST Lexicon (1994)• Syntactic, morphologic, orthographic information about

biomedical and general English terminology

Early Development of the UMLS Early Development of the UMLS Semantic Network (1988-1989)Semantic Network (1988-1989)

• UMLS collaborators asked to submit lists of useful semantic types and potential relationships between them- Active participation by BWH, Yale, Pittsburgh

• Purpose- Consistent categorization of all Metathesaurus

concepts

• Early attempts at organizing the suggested types into a network of interrelated types

First Released Version of UMLS First Released Version of UMLS Semantic Network (1990)Semantic Network (1990)

• 131 semantic types

- Each Metathesaurus concept assigned one or more semantic types, according to definitions of the types and a set of guidelines

• 35 relationships

- Relationships developed by top-down and bottom-up approaches and included definitions• Those deemed to be important for information retrieval

• Review of (implicit) relationships in MeSH and in MEDLINE citation records

Current Semantic NetworkCurrent Semantic Network

•135 semantic types

-2 major hierarchies• Entity

- Physical Object

- Conceptual Entity

• Event- Activity

- Phenomenon or Process

•54 relationships

Sample Semantic Type DefinitionSample Semantic Type Definition

UI: T190STY: Anatomical AbnormalityABR: anabSTN: A1.2.2DEF: An abnormal structure, or one that is abnormal in size or location.UN: Use this type if the abnormality in question can be either an acquired orcongenital abnormality. Neoplasms are not included here. These are given thetype 'Neoplastic Process'. If an anatomical abnormality has a pathologicmanifestation, then it will additionally be given the type 'Disease orSyndrome', e.g., "Diabetic Cataract" will be double-typed for this reason.HL: {isa} Anatomical Structure;{inverse_isa} Congenital Abnormality;{inverse_isa} Acquired Abnormality

Sample Relationship DefinitionSample Relationship Definition

UI: T151RL: affectsABR: AFRIN: affected_byRTN: R3.1DEF: Produces a direct effect on. Implied here is the altering or influencing ofan existing condition, state, situation, or entity. This includes has a role in,alters, influences, predisposes, catalyzes, stimulates, regulates, depresses,impedes, enhances, contributes to, leads to, and modifies.HL: {isa} functionally_related_to;{inverse_isa} interacts_with;{inverse_isa} disrupts;{inverse_isa} prevents …STL: [Anatomical Abnormality|Organism];

[Anatomical Abnormality|Physiologic Function] …

BodySystem

Body Spaceor Junction

Body Locationor Region

Entity

Physical Object Conceptual Entity

Substance Idea orConcept

FunctionalConcept

SpatialConcept

BodySubstance

EmbryonicStructure

Fully FormedAnatomicalStructure

Body Part, Organ orOrgan Component

Tissue Cell CellComponent

Gene orGenome

AnatomicalStructure

CongenitalAbnormality

AcquiredAbnormality

AnatomicalAbnormality

Portion of the Entity HierarchyPortion of the Entity Hierarchy

RelationshipsRelationships•Hierarchical (isa)

-Among types• Animal isa Organism

• Enzyme isa Biologically Active Substance

-Among relationships• treats isa affects

•Non-hierarchical (associative)

-Sign or Symptom diagnoses Pathologic Function

-Pharmacologic Substance treats Pathologic Function

Relationships (isa and associative)

A Portion of the Current Semantic Network

RelationshipsRelationships

•Relationship between a pair of semantic types is a possible link between the concepts assigned to those semantic types

-Relationship may or may not hold at the concept level

•A child semantic type inherits properties from its parents

Inheritance at Concept LevelInheritance at Concept LevelSemantic Network

Metathesaurus

AdrenalCortex

AdrenalCortical

hypofunction

Disease or SyndromeBody Part, Organ,

or Organ Component

Pathologic Functionisa

Biologic Function

isa

Fully FormedAnatomical

Structure

isa

location of

location of

Grouping SemanticTypesGrouping SemanticTypes

•Complexity of domain makes it difficult to

-Navigate and display the knowledge

-Reason with the objects in the domain

-Comprehend the conceptual space

•Semantic Network reduces the conceptual complexity of the UMLS, but

-For some purposes, smaller and coarser-grained groupings are needed

Semantic Type Groupings (2001)Semantic Type Groupings (2001)

•Clustered the larger set of semantic types into a small number of general groups

•Total of 15 groupings

•Effected an almost complete partitioning of the UMLS Metathesaurus

Grouping PrinciplesGrouping Principles

•Completeness

-Groups must cover the full domain

•Parsimony

-Number of groups should be as small as possible

•Naturalness

-Groups must be acceptable to a domain expert

Grouping Principles (cont.)Grouping Principles (cont.)

•Utility

-Groups must be useful for some purpose

•Semantic validity

-Groups must be semantically coherent

-Relationships shared by members of group

•Exclusivity

-Groups fully partition the domain

GroupingsGroupings (2001 Data) (2001 Data)

Some Relationships between Semantic GroupsSome Relationships between Semantic Groups

Distribution of Concepts in Distribution of Concepts in the UMLS (2001 Data)the UMLS (2001 Data)

Distribution of Concepts in Distribution of Concepts in PDQ (2001 Data)PDQ (2001 Data)

Research Applications of the Research Applications of the Semantic NetworkSemantic Network

• Natural language processing

• Information extraction and retrieval

• Ontological research

• Subsetting the domain

- E.g. extract all Metathesaurus concepts with a particular set of semantic types

• Conceptualizing the domain

- E.g., one resource oriented heavily to chemicals, another oriented to diseases

SummarySummary

•UMLS Semantic Network

-Provides overall conceptual structure to the UMLS by• Linking semantic types to Metathesaurus

concepts

• Providing a set of relationships to interrelate the types and (by inference) the concepts

• Allowing users to extract all concepts with a particular type

-Used in a number of research applications

-Variety of enhancements possible

Some ReferencesSome References

• McCray AT, Hole WT. The scope and structure of the first version of the UMLS Semantic Network. Proc Annu Symp Comput Appl Med Care, 1990; 126‑130.

• McCray AT. The UMLS Semantic Network. Proc Annu Symp Comput Appl Med Care. 1989; 503-7.

• McCray AT. Representing biomedical knowledge in the UMLS Semantic Network. High‑Performance Medical Libraries: Advances in Information Management for the Virtual Era. Westport: Meckler Publishing, 1993; 45‑55.

Some References (cont.)Some References (cont.)

• McCray AT, Nelson SJ. The representation of meaning in the UMLS. Methods Inf Med. 1995; 34(1‑2):193‑201.

• McCray AT, Burgun A, Bodenreider O. Aggregating UMLS semantic types for reducing conceptual complexity. MEDINFO. 2001; 216-220.

• McCray AT. An upper level ontology for the biomedical domain. Comp Funct Genom 2003; 4:80-4.

• Bodenreider O, McCray AT. Exploring semantic groups through visual approaches. Journal of Biomedical Informatics 2003;36(6):414-432.