Keynote at the AI in Medicine Conference (AIME 2005), giving an overview of the work in Ontology Mapping to people in Medical Informatics (which includes explaining the what and why of ontologies in general).

1. Ontology mapping:
a way out of
the medical tower of Babel?
Frank van Harmelen
Vrije Universiteit Amsterdam
The Netherlands Antilles

2. Before we start…
 a talk on ontology mappings
is difficult talk to give:
 no concensus in the field
• on merits of the different approaches
• on classifying the different approaches
 no one can speak with authority on
the solution
 this is a personal view, with a sell-by date
 other speakers will entirely disagree
(or disapprove)

3. Good overviews of the topic
 Knowledge Web D2.2.3:
“State of the art on ontology alignment”
 Ontology Mapping Survey
talk by Siyamed Seyhmus SINIR
 ESWC'05 Tutorial on
Schema and Ontology Matching
by Pavel Shvaiko Jerome Euzenat
 KER 2003 paper Kalfoglou & Schorlemmer
 These are all different & incompatible…

4. Ontology mapping:
a way out of
the medical tower of Babel?

5. The Medical tower of Babel
 Mesh
• Medical Subject Headings, National Library of Medicine
• 22.000 descriptions
 EMTREE
• Commercial Elsevier, Drugs and diseases
• 45.000 terms, 190.000 synonyms
 UMLS
• Integrates 100 different vocabularies
 SNOMED
• 200.000 concepts, College of American Pathologists
 Gene Ontology
• 15.000 terms in molecular biology
 NCI Cancer Ontology:
• 17,000 classes (about 1M definitions),

6. Ontology mapping:
a way out of
the medical tower of Babel?

7. What are ontologies &
what are they used for
world
concept
language
Agree on a
no shared understanding conceptualization
Conceptual and
terminological confusion Make it explicit
in some language.
Actors: both humans and machines

8. Ontologies come in very
different kinds
 From lightweight to heavyweight:
• Yahoo topic hierarchy
• Open directory (400.000 general categories)
• Cyc, 300.000 axioms
 From very specific to very general
• METAR code (weather conditions at air terminals)
• SNOMED (medical concepts)
• Cyc (common sense knowledge)

9. What’s inside an ontology?
 terms + specialisation hierarchy
 classes + class-hierarchy
 instances
 slots/values
 inheritance (multiple? defaults?)
 restrictions on slots (type, cardinality)
 properties of slots (symm., trans., …)
 relations between classes (disjoint, covers)
 reasoning tasks: classification, subsumption
Increasing semantic “weight”

10. In short
(for the duration of this talk)
 Ontologies are not
definitive descriptions of
what exists in the world (= philosphy)
 Ontologies are
models of the world
constructed
to facilitate communication
 Yes, ontologies exist
(because we build them)

11. Ontology mapping:
a way out of
the medical tower of Babel?

12.  Ontology mapping is
old & inevitable
 Ontology mapping is old
• db schema integration
• federated databases
 Ontology mapping is inevitable
• ontology language is standardised,
• don't even try to standardise contents

13.  Ontology mapping is
important
 database integration,
heterogeneous database retrieval
(traditional)
 catalog matching (e-commerce)
 agent communication (theory only)
 web service integration (urgent)
 P2P information sharing (emerging)
 personalisation (emerging)

14.  Ontology mapping is
now urgent
 Ontology mapping has acquired
new urgency
• physical and syntactic integration is ± solved,
(open world, web)
• automated mappings are now required (P2P)
• shift from off-line to run-time matching
 Ontology mapping has new opportunities
• larger volumes of data
• richer schemas (relational vs. ontology)
• applications where partial mappings work

15. Different aspects
of ontology mapping
 how to discover a mapping
 how to represent a mapping
• subset/equal/disjoint/overlap/
is-somehow-related-to
• logical/equational/category-theoretical
 atomic/complex arguments,
 confidence measure
 how to use it
We only talk about “how to discover”

16. Many experimental systems:
(non-exhaustive!)
 Prompt (Stanford SMI)  Coma (ULeipzig)
 Anchor-Prompt (Stanford SMI)  Buster (UBremen)
 Chimerae (Stanford KSL)  MULTIKAT (INRIA S.A.)
 Rondo (Stanford U./ULeipzig)  ASCO (INRIA S.A.)
 MoA (ETRI)  OLA (INRIA R.A.)
 Cupid (Microsoft research)  Dogma's Methodology
 Glue (Uof Washington)  ArtGen (Stanford U.)
 FCA-merge (UKarlsruhe)  Alimo (ITI-CERTH)
 IF-Map  Bibster (UKarlruhe)
 Artemis (UMilano)  QOM (UKarlsruhe)
 T-tree (INRIA Rhone-Alpes)  KILT (INRIA LORRAINE)
 S-MATCH (UTrento)

17. Different approaches to
ontology matching
 Linguistics & structure
 Shared vocabulary
 Instance-based matching
 Shared background knowledge

18. Linguistic &
structural mappings
 normalisation
(case,blanks,digits,diacritics)
 lemmatization, N-grams,
edit-distance, Hamming distance,
 distance = fraction of common parents
 elements are similar if
their parents/children/siblings are similar
decreasing order of boredom

19. Different approaches to
ontology matching
 Linguistics & structure
 Shared vocabulary
 Instance-based matching
 Shared background knowledge

20. Matching through
shared vocabulary
Q
Low(Q) Q Up(Q)
 Low(Q) µ Q µ  Up(Q)

21. Matching through
shared vocabulary
 Used in mapping geospatial databases
from German land-registration authorities
(small)
 Used in mapping bio-medical and
genetic thesauri
(large)

22. Different approaches to
ontology matching
 Linguistics & structure
 Shared vocabulary
 Instance-based matching
 Shared background knowledge

23. Matching through
shared instances

24. Matching through
shared instances
 Used by Ichise et al (IJCAI’03) to
succesfully map parts of Yahoo to
parts of Google
 Yahoo = 8402 classes, 45.000 instances
 Google = 8343 classes, 82.000 instances
 Only 6000 shared instances
 70% - 80% accuracy obtained (!)
 Conclusions from authors:
• semantics is needed to improve on this ceiling

25. Different approaches to
ontology matching
 Linguistics & structure
 Shared vocabulary
 Instance-based matching
 Shared background knowledge

26. Matching using shared
background knowledge
shared
background
knowledge
ontology 1 ontology 2

27. Ontology mapping
using background knowledge
Case study 1
PHILIPS Work with Zharko Aleksovski @ Philips
• Michel Klein @ VU
KIK @ AMC

28. Overview of test data
Two terminologies from
intensive care domain
 OLVG list
• List of reasons for ICU admission
 AMC list
• List of reasons for ICU admission
 DICE hierarchy
• Additional hierarchical knowledge describing
the reasons for ICU admission

29. OLVG list
 developed by clinician
 3000 reasons for ICU admission
 1390 used in first 24 hours of stay
• 3600 patients since 2000
 based on ICD9 + additional material
 List of problems for patient admission
 Each reason for admission is described with
one label
• Labels consist of 1.8 words on average
• redundancy because of spelling mistakes
• implicit hierarchy (e.g. many fractures)

30. AMC list
 List of 1460 problems for ICU admission
 Each problem is described using
5 aspects from the DICE terminology:
 2500 concepts (5000 terms), 4500 links
• Abnormality (size: 85)
• Action taken (size: 55)
• Body system (size: 13)
• Location (size: 1512)
• Cause (size: 255)
 expressed in OWL
 allows for subsumption & part-of reasoning

31. Why mapping
AMC list $ OLVG list?
 allow easy entering of OLVG data
 re-use of data in
• epidemiology
• quality of care assessment
• data-mining (patient prognosis)

32. Linguistic mapping:
 Compare each pair of concepts
 Use labels and synonyms of concepts
 Heuristic method to discover
equivalence and subclass relations
Long brain tumor More specific Long tumor
than
 First round
• compare with complete DICE
• 313 suggested matches, around 70 % correct
 Second round:
• only compare with “reasons for admission” subtree
• 209 suggested matches, around 90 % correct
 High precision, low recall (“the easy cases”)

33. Using background knowledge
 Use properties of concepts
 Use other ontologies to discover
relation between properties
?
…. ….
…. ….
…. ….

34. Semantic match
DICE aspect
Lexical match taxonomies Given
? Abnormality taxonomy
? Action taxonomy
? Body system taxonomy
? Location taxonomy
? Cause taxonomy
Implicit
OLVG matching: DICE
problem list property problem list
match

35. Semantic match
Taxonomy of body parts
Blood vessel
is more general is more general
Vein
Artery
is more general
Aorta
Lexical match: Lexical match:
has location Reasoning: has location
implies
Aorta thoracalis dissection Dissection of artery
Location match:
has more
general location

36. Example: “Heroin intoxication”
– “drugs overdose”
Cause taxonomy
Drugs
is more general
Heroine
Lexical
match: Lexical
Cause match: match:
cause
has more specific cause
cause
Heroin intoxication
Drugs overdosis
Abnormality match:
has more general
Lexical abnormality Lexical match:
match: abnormality
abnormality Abnormality taxonomy
Intoxicatie
is more general
Overdosis

37. Example results
• OLVG: Acute respiratory failure abnormality
DICE: Asthma cardiale
• OLVG: Aspergillus fumigatus cause
DICE: Aspergilloom
• OLVG: duodenum perforation abnormality,
DICE: Gut perforation cause
• OLVG: HIV
cause
DICE: AIDS
• OLVG: Aorta thoracalis dissectie type B location,
DICE: Dissection of artery abnormality

38. Extension:
approximate matching
 Terms are not precisely defined
 Terms are not precisely used
Exact reasoning will not be useful
A B A½B?

39. Approximate matching
 Translate every class-name into a
propositional formula
(both DNF and CNF versions)
A ⊆ B = (∪Ai ⊆ ∩Bk) = ∀i,k (Ai ⊆ Bk)
 ignore increasing number. of
(i,k)-subsumption pairs
 varies from classical to trivial

40. Results
(obtained on different domain)
600000
500000
400000
B subClass of A
300000 A subClass of B
equivalences
200000
100000
0
0
3
4
5
6
8
9
1
2
7
0
0.
0.
0.
1.
0.
0.
0.
0.
0.
0.
0.

41. Ontology mapping
using background knowledge
Case study 2
Work with Heiner Stuckenschmidt
@ VU

42. Case Study:
 Map GALEN & Tambis,
using UMLS as background knowledge
 Select three topics with sufficient overlap
• Substances
• Structures
• Processes
 Define some
partial & ad-hoc manual mappings
between individual concepts
 Represent mappings in C-OWL
 Use semantics of C-OWL
to verify and complete mappings

43. Case Study:
verification & verification &
derivation UMLS derivation
(medical terminology)
lexical mapping lexical mapping
GALEN Tambis
(medical ontology) derived mapping (genetic ontology)

44. Ad hoc mappings: Substances
UMLS GALEN
Notice: mappings high and low in the hierarchy, few in the middle

45. Ad hoc mappings: Substances
UMLS Tambis
Notice different grainsize: UMLS course, Tambis fine

46. Verification of mappings
=
UMLS:Chemicals
UMLS:Chemicals_ Tambis:Chemical
viewed_structurally
Tambis:enzyme
⊥?
UMLS:Chemicals_
viewed_functionally
UMLS:enzyme
=

47. Deriving new mappings
UMLS:substance UMLS:Phenomenon_
or_process
UMLS:Chemicals ⊥
Galen:
ChemicalSubstance
UMLS:OrganicChemical
⊇
=
⊆

48. Ontology mapping:
a way out of
the medical tower of Babel?

49. “Conclusions”
 Ontology mapping is (still) hard & open
 Many different approaches will be required:
• linguistic,
• structural
• statistical
• semantic
• …
 Currently no roadmap theory on
what's good for which problems

50. Challenges
 roadmap theory
 run-time matching
 “good-enough” matches
 large scale evaluation methodology
 hybrid matchers (needs roadmap theory)

51. Ontology mapping:
a way out of
the medical tower of Babel?
?