Vipul Kashyap, National Library of Medicine kashyapnlm.nih.gov

1

The UMLS? Semantic Network and the
Semantic Web

• Vipul Kashyap, National Library of
  Medicinekashyap_at_nlm.nih.gov
• November 11, 2003
• AMIA 2003 Annual Symposium
• Washington, DC

2
Outline

• The Unified Medical Language System (UMLS)
• The UMLS? Semantic Network
• Motivation Formal Representations of Biomedical
  Knowledge
• DAMLOIL/OWL An Ontology Language for the
  Semantic Web
• Vanilla Representation of the Semantic Network
• Requirements of the Semantic Network
• Blocking Inheritance
• Ad-hoc Polymorphisms
• Conclusions and Future Work

3
The Unified Medical Language System (UMLS)
Anatomical Structure
Semantic Types in theSemantic Network
Disease or Syndrome
Population Group
Pharmacologic Substance
Body Part, Organ or Organ Component
Metathesaurus Concepts
Saccular Viscus
Medias-tinum
Angina Pectoris
Esophagus
Heart
Cardiotonic Agents
Left PhrenicNerve
Tissue Donors
Heart Valves
Fetal Heart
Thesauri/Vocabulary/ Classifications

4
The UMLS Semantic Network (Partial View)
5
Motivation Formal Representations of Biomedical
Knowledge

• Improved Specification of Queries for Medical
  Information Retrieval
• (Mumps/CO AND Pancreatitis/ET)
  ?complication.Mumps ? ?etiology.Pancreatitis
• Precise specifications of MeSH expressions using
  OWL concepts
• Use of DL-type inferences to improve search
  results
• Semantic Vocabulary Interoperation Project
• http//cgsb2.nlm.nih.gov/kashyap/projects/SVIP
• Translations of concepts across multiple
  biomedical vocabularies
• Inferring concept equivalence and inconsistency
  for vocabulary merging
• Determination of semantic position of a concept
  in a DAG structure
• Subsumption checking
• Consistency of biomedical terminologies

6
DAMLOIL/OWL An Ontology Language for the
Semantic Web

• Describes the structure of a domain
• Classes and properties
• Axioms assert characteristics of these classes
  and properties
• A very expressive Description Logic (DL)
• Classes Names (URIs) or Expressions
• Class and property constructors for building
  Class Expressions
• unionOf, intersectionOf, complementOf, one of
• toClass, hasClass, minCardinalityQ,
  maxCardinality
• Axioms
• Assert subsumption or equivalence of classes
  (names or expressions)
• Assert disjointness or non-equivalence of classes
  (names or expressions)
• Assert domains and ranges of properties

7
DAMLOIL/OWL Notation

• DL Notation Bacterium ? Virus
• OWL NotationltowlClassgt ltowlunionOf
  rdfparseTypeCollectiongt ltowlClass
  rdfaboutBacterium/gt ltowlClass
  rdfaboutVirus/gt lt/owlunionOfgtlt/owlCl
  assgt
• Domain/Range Notationdomain(process_of)
  BiologicFunctionrange(process_of) Organism
• OWL NotationltowlObjectProperty
  rdfIDprocess_ofgt ltrdfsdomain
  rdfresourceBiologicFunctiongt
  ltrdfsrange rdfresourceOrganismgtlt/owlObjec
  tPropertygt

8
DAMLOIL/OWL Constructs
9
Vanilla Representation using DAMLOIL/OWL

• Semantic Types ? DAMLOIL/OWL Classes
• Fungus ? Organism
• Virus ? Organism
• Semantic Relationships ? OWL properties
• part_of ? physically_related_to
• contains ? physically_related_to
• Domains and Ranges of Relationships Later
• Properties of Semantic Network Relationships
• Asymmetric relationships
• has_part part_of?
• Symmetric relationships
• adjacent_to adjacent_to?

10
Requirements of the Semantic Network

• Consider the association
• process_of(BiologicFunction, Organism)
• Domain inheritance
• process_of is inherited by descendants of
  BiologicFunction
• E.g., process_of(MentalProcess, Organism)
• Range inheritance
• process_of is inherited to descendants of
  Organism
• E.g., process_of(BiologicFunction, Plant)
• Inheritance blocking
• process_of(MentalProcess, Plant) is inherited
  from process_of(BiologicFunction, Organism)
• Undesirable inference, needs to be blocked !
• Ad-hoc polymorphism
• Multiple domains or ranges of associations
• Multiple domains (or ranges) not related by is-a
  relationship (NOT sub-type polymorphism)

11
Blocking Inheritance Domain Blocking

• domain(process_of) BiologicFunctionrange(proces
  s_of) Organism
• To prevent inference of process_of(MentalProcess,
  Plant)
• Block the inheritance of process_of from
  BiologicFunction to MentalProcess
• domain(process_of) BiologicFunction ?
  ?MentalProcess
• ?process_of.T ? BiologicFunction ?
  ?MentalProcess
• MentalProcess ? ? 0 process_of
• MentalProcess ? ? 0 process_of Plant

12
Blocking Inheritance Range Blocking

• domain(process_of) BiologicFunctionrange(proces
  s_of) Organism
• To prevent inference of process_of(MentalProcess,
  Plant)
• Block the inheritance of process_of from Organism
  to Plant
• range(process_of) Organism ? ?Plant
• T ? ?process_of.Organism ? ?Plant
• Plant ? ? 0 process_of?
• Plant ? ? 0 has_process
• Plant ? ? 0 has_process MentalProcess

13
Ad-hoc Polymorphism

• Consider the following associations
• contains(BodySpaceOrJunction,
  BodyPartOrganOrOrganComponent)contains(EmbryonicS
  tructure, BodySubstance)
• IFdomain(contains) BodySpaceOrJunction ?
  EmbryonicStructurerange(contains)
  BodyPartOrganOrOrganComponent ? BodySubstance
• THEN, we might miss contains(x, y), wherex ?
  BodySpaceOrJunction ? ?EmbryonicStructurey ?
  BodyPartOrganOrOrganComponent ? ?BodySubstance
• AND IF the intersections are empty, we may
  even get an empty association!
• IFdomain(contains) BodySpaceOrJunction ?
  EmbryonicStructurerange(contains)
  BodyPartOrganOrOrganComponent ? BodySubstance
• THEN we might introduce spurious contains(x,y)
  wherex ? BodySpaceOrJunction, y ? BodySubstance
  OR
• x ? EmbryonicStructure, y ? BodyPartOrganorOrga
  nComponent

14
Ad-hoc Polymorphism Domain/Range Factorization

• Consider
• analyzes(DiagnosticProcedure,
  BodySubstance)analyzes(LaboratoryProcedure,
  BodySubstance)analyzes(DiagnosticProcedure,
  Chemical)analyzes(LaboratoryProcedure, Chemical)
• Each domain type is associated with each range
  type!
• domain(analyzes) DiagnosticProcedure ?
  LaboratoryProcedure
• range(analyzes) BodySubstance ? Chemical

15
Ad-hoc Polymorphism Property Renaming

• define contains1domain(contains1)
  BodySpaceOrJunctionrange(contains1)
  BodyPartOrganOrOrganComponent
• define contains2domain(contains2)
  EmbryonicStructurerange(contains2)
  BodySubstance
• Assert the following axioms
• contains1 ? contains contains2 ? contains
  contains ? contains1 ? contains2

16
Adhoc Polymorphism Property Restrictions

• BodySpaceJunction ? ?contains.BodyPartOrganOrOrgan
  Component
• EmbryonicStructure ? ?contains.BodySubstance
• For any class C, that is not a subclass of
  BodySpaceJunction or EmbryonicStructure
• C ? ? 0 contains
• Questions
• Does it mean that other classes are not allowed
  to participate in the relationship?
• What about other interpretations?
• EmbryonicStructure ? ?contains.BodySubstance ?
• There is at least some EmbryonicStructures
  related to BodySubstances via the contains
  association

17
Interesting Questions and Issues

• Can/should something be inferred from absence of
  a link?
• A ? ?P.B doesnt prohibit ?A from being related
  to B
• Should relationships be inferred to be asymmetric
  by default?
• Automatically assert ?(P ? P?)
• Are the is-a children disjoint?
• Untangling of ontologies
• Is it important to detect inconsistency in the
  ontology?
• Are P(A, B) and P(A, C) inconsistent if B ? C ? ?
  ?
• A ? ?P.B will not detect inconsistency unless we
  assert A ? ?P.T
• Limit the types of axioms if questions asked by
  an application do not require it.
• E.g. axioms regarding asymmetric relationships
• Are there approximations to concepts and axioms?
• Assert P1 ? P, P2 ? P instead of P ? P1 ? P2

18
Conclusions and Future Work

• Have presented experiences in representing a real
  world ontology using a widely publicized
  standard DAMLOIL/OWL
• Caution It is not as straightforward as it seems
• Identified the requirements of the UMLS? Semantic
  Network
• Investigated the various alternatives in
  DAMLOIL/OWL to capture these requirements
• Discussed issues and criteria that need to be
  considered for representing bio-medical knowledge
  using DAMLOIL/OWL
• Future Work
• Investigate representation of UMLS Metathesaurus
  using DAMLOIL/OWL
• Issues related to representation of BT/NT or
  Parent/Child Relationships
• Use of DAMLOIL/OWL
• Semantic Vocabulary Interoperability
• Enhanced Information Retrieval