swanson rfo-pres-grp-mtg
DESCRIPTION
Presentation to Kno.e.sis group on graph-based techniques for decomposing Swanson's RS-DFO HypothesisTRANSCRIPT
A Graph-Based Decomposition of Swanson’s Hypothesis using
Semantic Predications
Delroy Cameron, Olivier Bodenreider, Hima Yalamanchili, Tu Danh, Sreeram Vallabhaneni, Krishnaprasad Thirunarayan,
Amit P. Sheth and Thomas C. Rindflesch
02/17/2012
2
OUTLINEBackground
Swanson’s HypothesisLiterature-Based Discovery (LBD)
Problem
Graph-based Approach
Experimental Results
Future Work
Conclusion
3
SWANSON’S HYPOTHESIS
Source: http://academic.research.microsoft.com/VisualExplorer#185977
4
SWANSON’S HYPOTHESISRaynaud Syndrome – Fish Oil Hypothesis
(1986)1
Scientific Literatureest. 1000 Fish Oil articlesest. 2000 Raynaud articles
FeaturesFew co-referencesOverlapping terms/concepts
“dietary fish oil might prevent Raynaud’s syndrome.”
2 relevant articles, S. Moncada
63 relevant papers
25 DFO 34 RS4
489 Medline + Embase
4
DFO RS
1Swanson D. Fish oil, raynaud’s syndrome, and undiscovered public knowledge. Perspect Biol Med 1986;30(1):7–18.
5
SWANSON’S HYPOTHESIS
Platelet Aggregatio
n
DFOBlood
ViscosityRaynaud
Syndrome
Vascular Reactivity
“dietary fish oil might prevent Raynaud’s syndrome.”
INHIBITS
INHIBITS
INHIBITS
CAUSES
CAUSES
CAUSES
6
LITERATURE-BASED DISCOVERY(LBD)
Finding explicit connectionsUndiscovered public knowledge2
Noninteracting literatures
ABC Model
2Swanson DR.Undiscovered public knowledge.Library Quarterly 1986;56(1):103–18.
A B C
7
PROBLEMIR-based techniques
Lack context
ABC ModelLack coverageAnC3 Model, n={B1, B2…Bm}
3Wilkowski B, Fiszman M, Miller C, Hristovski D, Arabandi S, Rosemblat G, Rindflesch T. Discovery browsing with semantic predications and graph theory. AMIA Annu Symp Proc 2011
DFO
Platelet Aggregatio
n
RaynaudSyndro
meINHIBITS CAUSE
S
Epoprostenol
Prostaglandin
STIMULATES ISA
INHIBITS
A C
B1
B2
B3B
8
PROBLEM-SOLUTION AnC Model
Multiple Paths3
Subgraphs3
Background knowledge*
DFO
RaynaudSyndromeTREATS
Epoprostenol
STIMULATES
DFO
Platelet Aggregatio
n
RaynaudSyndrom
e
INHIBITS
CAUSES
Prostaglandin
PRODUCES
ISASTIMULATES TREATS
3Wilkowski B, Fiszman M, Miller C, Hristovski D, Arabandi S, Rosemblat G, Rindflesch T. Discovery browsing with semantic predications and graph theory. AMIA Annu Symp Proc 2011
9
CONTRIBUTIONGraph-based Framework for LBD
Semantic predications From ABC Model to AnC Model Construct expressive subgraphs Background knowledge*
Evaluate/Assess Recover Swanson’s Hypothesis Decompose Swanson’s Hypothesis*
Question Answering
Information Retrieval
Literature-based Discovery
10
ABC Mode
l
AnC Mode
l
Graph Theor
y
Subgraphs
Background
knowledge
Hristovski [14,15,26] (BITOLA)
✔ ✔
Pratt [13, 22](LitLinker)
✔ ✔
Ahlers [16] ✔
Cohen [23](Epiphanet)
✔ ✔
Wilkowski [17] ✔ ✔ ✔
Cameron* ✔ ✔ ✔ ✔
RELATED WORK (Semantics-based)
11
IMPACTInterdisciplinary Research
Mechanisms of operationCausality Relationships
Semantic Integration of Heterogeneous DataExperimental DataScientific LiteratureStructured Data
12
1. Literature Preprocessing
2. Predication Extraction (SemRep)
3. Predications Graph Creation
4. Path generation Extend ABC Model to AnC Model
5. Subgraph Construction Multiple Paths Background Knowledge
APPROACH
Path Generation
Corpus
13
RS
Path GenerationReachability
“the notion of being able to get from one vertex in a directed graph to some other vertex”
A
C
DFO ReachabilityRelation
Transitive Closure =
14
Subgraph CreationOriginal Swanson Associations
Thoroughly read Paper1
Three (3) Primary Associationso Eight (8) Supplementary Associationso Eight (8) Secondary Associations
Subgraph Construction (manual)Classifying Paths from Reachability RelationHuman Cognition as Background Knowledge
1Swanson D. Fish oil, raynaud’s syndrome, and undiscovered public knowledge. Perspect Biol Med 1986;30(1):7–18.
15
Primary/Supplementary Associations
16
Dataset & ExperimentsExperiment I
Baseline1: Full Text (Titles, Abstracts, Text)
Experiment IIBaseline2: Titles, Abstracts
17
Experiments
Reachability Relation= 36 paths
Reachability Relation= 146 paths
Platelet Aggregation (path#1) Primary
Dietary Fish Oil->INHIBITS->platelet aggregation->CAUSES->Raynaud Syndrome
Epoprostenol
Raynaud Syndrome
DISRUPTS
DISRUPTS
Platelet AggregationDFO CAUSES
STIMULATES TREATS
Four Relevant Paths
Platelet Aggregation (path#2) Supplementary
DFO
Dietary Fish Oil->PRODUCES->Prostaglandin (PGI3)-> INHIBITS->platelet aggregation->CAUSES->Raynaud Syndrome
Epoprostenol
Raynaud Syndrome
Platelet Aggregation
Prostaglandin I3
ISA
CONVERTS_TO
DISRUPTS CAUSES
Prostaglandin
ISACONVERTS_TO
DISRUPTS
DISRUPTS
STIMULATES TREATS
DISRUPTS
Eight Relevant Paths
TREATS
TREATS
Platelet Aggregation (path#3) PrimaryDietary Fish Oil->INHIBITS->Blood Viscosity->CAUSES ->Raynaud Syndrome
Raynaud Syndrome
Ketanserin
TREATSDISRUPTS
Epoprostenol
TREATS
Blood Viscosity
DISRUPTS
STIMULATES DISRUPTS
CAUSESDFO
Four Relevant Paths
Platelet Aggregation (path#4) SupplementaryDietary Fish Oil->INHIBITS->triglyceride->ISA->Blood Lipid->AFFECTS->Blood Viscosity->CAUSES ->Raynaud Syndrome
Raynaud SyndromeDFO Blood
ViscosityDISRUPTS CAUSES
Fatty Acid
ISA
Essential Fatty Acid
ISA
Triglyceride
INHIBIT
AFFECTS
Lipid
ISA
Six Relevant Paths
INHIBITS
22
Experimental Results
Association Type
Experiment I Experiment II
Primary (3) 3 2* (1VR missing)
Supplementary (8)
4 (2VR, 2BV missing)
4* (2VR, 2BV missing)
Secondary (8) 7 (1VR missing) 5* (3VR missing)
* - Not sufficiently detailed
23
ABC Mode
l
AnC Mode
l
Graph Theor
y
Subgraphs
Background
knowledge
Hristovski [14,15,26] (BITOLA)
✔ ✔
Pratt [13, 22](LitLinker)
✔ ✔
Ahlers [16] ✔
Cohen [23](Epiphanet)
✔ ✔
Wilkowski [17] ✔ ✔ ✔
Cameron* ✔ ✔ ✔ ✔
RELATED WORK (Semantics-based)
24
FUTURE WORKScalability
Path Clusteringo Structural Features (Centrality, Geodesic)o Semantic Features (Similarity, Relatedness)
Semantic Integration of Background Knowledge
o Heuristics (knowledge abstraction,[CameronBIBM2011])
Inconsistencies
Predication Extraction (SemRep)
Corpus (Richness)
25
CONCLUSIONGraph-based Framework (GFB) for
LBDSemantic predicationsExtend ABC Model to AnC ModelConstruct expressive subgraphsBackground knowledge*
Evaluate/AssessRecover Swanson’s HypothesisDecompose Swanson’s Hypothesis*
The future of LBD requires the
Semantic Integration of
subgraphs and background knowledge
27
POTENTIAL DISCOVERY
Raynaud Syndrome
AspirinEpoproste
nolINHIBITS TREATS
28
QUESTIONS