Textbased Discovery in Biomedicine The Architecture of the DADsystem

1
Text-based Discovery in BiomedicineThe
Architecture of the DAD-system

• Marc Weeber1,2, Henny Klein1,
• Alan R. Aronson2, Jim G. Mork2,
• Lolkje T. W. de Jong - van den Berg1,
• Rein Vos1,3

1Department of Social Pharmacy and
Pharmacoepidemiology, Groningen University
Institute for Drug Exploration, The
Netherlands 2Lister Hill National Center for
Biomedical Communication, National Library of
Medicine, Bethesda, MD 3Health Ethics and
Philosophy, Faculty of Health Sciences,
University of Maastricht, The Netherlands
2
Introduction

• Goal
• Finding new biomedical knowledge through the
  combination of existing knowledge as represented
  in the medical literature
• Motivation
• Prevention of re-inventing the wheel, re-usage of
  specific knowledge outside the original domain of
  discovery

3
Swanson
A
C
B
?

• AB Raynauds disease is characterized by high
  blood viscosity and high platelet aggregation
• BC Fish oil is known to reduce blood viscosity
  and platelet aggregation

4
Vos and Rikken

• Drugs instead of diet factors
• Intermediate (B) terms are adverse drug reactions
• Drug Adverse drug reactions Disease The
  DAD-system
• Vos (1991) Drugs looking for diseases

5
Existing Techniques

• Swanson Smalheiser
• Single words/multi word terms
• MEDLINE titles
• No statistics
• Gordon Lindsay
• Single words/multi word terms
• Information Retrieval statistics
• Replication of Swansons discoveries

6
New Techniques

• Use of UMLS concepts
• PubMed
• MetaMap mapping free text (MEDLINE titles and
  abstracts) to concepts
• Interactive web interface

7
Two-step Approach

• Open discovery, generating a hypothesis

• Closed discovery, testing a hypothesis

8
Why UMLS Concepts?

• Use of only biomedically relevant information
• Useful transition from single word to multi word
  term
• Semantic information (semantic types) for
  filtering (e.g. select only Disease or Syndrome)

9
DAD-system
KS
10
DAD-system
KS
Filter
Select
11
DAD-system
KS
Filter
Select
12
Open Discovery
A

• Query (user input)
• raynauds disease

13
Open Discovery
A

• Mapping text to concept through MetaMap
• Raynaud's Disease Disease or Syndrome

14
Open Discovery
A

• Synonym lookup
• Raynaud's syndrome
• Raynaud's disease /phenomenon
• Variant generation
• e.g. syndrome / syndromes

15
Open Discovery
A

• PubMed query
• raynaud OR raynauds
• Processing query in titles and abstracts
• Result 1,246 MEDLINE citations

16
Open Discovery
A

• Text to concept mapping of all citations
• Sentences with Raynauds disease
• Result 1,278 UMLS concepts

17
Open Discovery
A

• Select functional/physiological concepts
• Semantic types in filter
• Body Location or Region
• Biologic Function
• Cell Function
• Phenomenon or Process
• Physiologic Function
• Tissue

18
Open Discovery
A
B

• Result 57 Concepts
• Frequency range
• 1- 18

19
Open Discovery
A
B

• Selected B-concepts
• Plasma Viscosity Level
• Blood Viscosity
• Platelet Adhesiveness
• Platelet Aggregation
• Effects, Blood Coagulation

20
Open Discovery
A
B

• Variants
• plasma, plasmas
• viscosity, viscous,
• aggregation, aggregations, aggregating
• coagulation, coagulating

21
Open Discovery
A
B

• PubMed query
• blood coagulation OR blood viscosity OR plasma
  viscosity OR platelet adhesiveness OR platelet
  aggregation
• Result 10,611 MEDLINE citations

22
Open Discovery
A
B

• Concepts in sentences with B-concepts
• 7,702
• Concepts not in Raynaud sentences
• 6,747

23
Open Discovery
A
B

• Filter for dietary related concepts
• Semantic types in filter
• Vitamin
• Lipid
• Element, Ion, or Isotope

24
Open Discovery
A
B
C

• Result 206 Concepts
• Rank order on relations
• Fish oil related concepts

Eicosapentaenoic Acid Fish Oil Fatty Acids, Omega
3 MAXEPA Omega-3 Polyunsaturated Fatty Acid Cod
Liver Oil Salmon Oil
25
Closed Discovery
A
C
Eicosapentaenoic Acid Fish Oil Fatty Acids, Omega
3 MAXEPA Omega-3 Polyunsaturated Fatty Acid Cod
Liver Oil Salmon Oil
Raynauds Disease
26
Closed Discovery
A
C
1,246 citations 1,278 concepts
463 citations 1,795 concepts
479 common concepts
27
Closed Discovery
A
C
Functional / Physiological Filter
45 B-concepts
28
Closed Discovery
A
C
B

• New concepts
• Vasodilatation
• Veins, Capillaries
• Dinoprostone
• Fibrinolysis
• Deformability
• Rheology

• Known concepts
• Plasma viscosity level
• Blood Viscosity
• Platelet Adhesiveness
• Platelet Aggregation
• Effects, Blood
• Coagulation

29
Juxtaposition
30
Success / Failure

• Simulation of Raynauds disease fish oil and
  migraine magnesium
• Discovery of new therapeutic applications for
  thalidomide
• Mapping (Mg milligram / magnesium)
• Association defined by co-occurrence

31
Future

• Better semantic analysis
• increase(A,B) and decrease(B,C)
• Better user interface
• More databases
• e.g. finding genetic bases for diseases