Paper: http://jodischneider.com/pubs/argdiap2014.pdf
We present a motivating scenario for using argumentation within a scientific knowledge base. Our goal is to transform natural language papers, with the manual work of expert curators, into elaborated claim-argument networks.
In ongoing work, we are focusing on creating Micropublications (Tim Clark, Paolo N. Ciccarese, Carole A. Goble http://arxiv.org/abs/1305.3506 ) through manual annotation. Currently we are engaging with pharmaceutical curators to envision the most appropriate way to record evidence for the next generation of the the Drug Interaction Knowledge Base.
Presented at The 12th ArgDiaP Conference "From Real Data to Argument Mining"
https://sites.google.com/site/argdiapen/12th-argdiap
Emergent Methods: Multi-lingual narrative tracking in the news - real-time ex...
Modeling arguments in scientific papers ArgDiaP 2014 05-23
1. Jodi Schneider, Carol Collins, Lisa E Hines,
John R Horn, and Richard Boyce
12th Argumentation, Dialogue, Persuasion
conference (ArgDiaP 2014)
Warsaw, Poland
2014-05-25
6. Goal: Support evidence-based updates
to drug-interaction reference DBs
• Make sense of the EVIDENCE
– New clinical trials
– Adverse drug event reports
– Drug product labels
– Updates to regulatory
information (U.S. FDA,…)
– …
• Significant discrepancies between different
drug-interaction reference DBs
http://jama.jamanetwork.com/article.aspx?articleid=183454
7. Drug Interaction Knowledge Base
(DIKB) - Boyce 2007-2009
– Hand-constructed knowledge base
– Safety issues when 2 drugs are taken together
– Focus is on EVIDENCE
8. Drug Interaction Knowledge Base
(DIKB) - Boyce 2007-2009
– Hand-constructed knowledge base
– Safety issues when 2 drugs are taken together
– Focus is on EVIDENCE
All assumptions are linked to evidence
Enables the system to identify when
9. DIKB supports queries about
assertions & evidence:
• Get all assertions that are supported by a
U.S. FDA regulatory guidance statement
• Are the evidence use assumptions are
concordant, unique, and non-ambiguous?
• Which assertions are supported/refuted by
just one type of evidence?
10. Limitations of DIKB v1.2
• Minimal argumentation model
– swanco:citesAsSupportingEvidence
– swanco:citesAsRefutingEvidence
• Cannot recover the source text
– Document-level citation
– Quote & section citation preferrable
• Level of detail
– Want more detail on data, methods, materials
11. Record deeper relationships between
assertions and evidence
• Assertions
– “there (is/is not) an interaction between A & B”
– “Enzyme E reduces clearance of drug D by 25%”
• Evidence
– Scientific literature
– Drug product labeling
– U.S. FDA guidance documents
12.
13. Micropublication: Claim + Support
(e.g. Attribution)
Micropublications: a Semantic Model for Claims, Evidence, Arguments and Annotations in Biomedical Communications
Tim Clark, Paolo N. Ciccarese, Carole A. Goble
http://arxiv.org/abs/1305.3506
14. Constructs claim-argument network
across scientific papers
Micropublications: a Semantic Model for Claims, Evidence, Arguments and Annotations in Biomedical Communications
Tim Clark, Paolo N. Ciccarese, Carole A. Goble
http://arxiv.org/abs/1305.3506
16. Micropublications Ontology
Micropublications: a Semantic Model for Claims, Evidence, Arguments and Annotations in Biomedical Communications
Tim Clark, Paolo N. Ciccarese, Carole A. Goble
http://arxiv.org/abs/1305.3506
17. What does this do for the DIKB
• Network of claims with supports & challenges
• Can record materials, methods, data
• Quotes can be naturally linked into the graph
• We can query for every mp:Claim that has
no support.
– An assumption (mp:Claim)
– Can have its own support graph, specified once
– We can query the support graph
24. From individual documents to a
searchable claim-argument network
• "Pay as you go" annotation of source
documents with Domeo & Micropublications
• Generates claim-argument network
– Supports & challenges
– Materials, methods, data
– Quotes linked into the graph
– … within & across documents
• Query support
Notes de l'éditeur
https://sites.google.com/site/argdiapen/12th-argdiap
Paper online:
http://jodischneider.com/pubs/argdiap2014.pdf
Image from http://www.clipartbest.com/clipart-McLLpbGKi
Adverse drug events are a leading cause of death
Image from https://www.njpharmacy.com/wp-content/uploads/2013/02/drug-interactions-checker.png
Image from http://www.clipartbest.com/clipart-McLLpbGKi
Adverse drug events are a leading cause of death
Images from
http://www.knowabouthealth.com/android-version-of-medscape-app-ready-to-download/7568/
Android Play store
http://amazingsgs.blogspot.com/2011/10/top-5-free-android-medical-apps-for.html
Adverse drug events are a leading cause of death
Images from
http://www.knowabouthealth.com/android-version-of-medscape-app-ready-to-download/7568/
Android Play store
http://amazingsgs.blogspot.com/2011/10/top-5-free-android-medical-apps-for.html
Image from http://jama.jamanetwork.com/article.aspx?articleid=183454
# How many pharmacokinetic studies in the DIKB could be used to support or refute
# an increases AUC assertion?
# How many AUC studies are in the DIKB that are based on data from the
# product label?
# Are the evidence use assumptions are concordant, unique, and non-ambiguous
# Get all assertions that are supported by an FDA guidance statement
# SHOW THE DISTRIBUTION OF THE LEVELS OF EVIDENCE FOR MECHANISTIC ASSERTIONS
# WHAT ASSERTIONS ARE SUPPORTED/REFUTED BY JUST ONE TYPE OF EVIDENCE?
## which items in the DIKB have evidence for and against that is both
## from product labeling?
# number of assertions in the system
# number of evidence items for and against
## find assertions that are not supported by evidence
# what single evidence items act as as support or rebuttal for multiple substrate_of assertions?
#remove a older version of a redundant evidence item
## change evidence for to evidence against
# has this evidence item been rejected
# what other assertions are being supported/challeged by this evidnece item?
# How many pharmacokinetic studies in the DIKB could be used to support or refute
# an increases AUC assertion?
# How many AUC studies are in the DIKB that are based on data from the
# product label?
# Are the evidence use assumptions are concordant, unique, and non-ambiguous
# Get all assertions that are supported by an FDA guidance statement
# SHOW THE DISTRIBUTION OF THE LEVELS OF EVIDENCE FOR MECHANISTIC ASSERTIONS
# WHAT ASSERTIONS ARE SUPPORTED/REFUTED BY JUST ONE TYPE OF EVIDENCE?
## which items in the DIKB have evidence for and against that is both
## from product labeling?
# number of assertions in the system
# number of evidence items for and against
## find assertions that are not supported by evidence
# what single evidence items act as as support or rebuttal for multiple substrate_of assertions?
#remove a older version of a redundant evidence item
## change evidence for to evidence against
# has this evidence item been rejected
# what other assertions are being supported/challeged by this evidnece item?
Micropublications: a Semantic Model for Claims, Evidence, Arguments and Annotations in Biomedical Communications
Tim Clark, Paolo N. Ciccarese, Carole A. Goble
http://arxiv.org/abs/1305.3506
What is wrong with this model?
Limited automatic reasoning
Limited integration with annotation
What are the new opportunities?
New model by the authors of SWAN/SIOC
Micropublications enable explicitly representing the warrant, i.e. the materials and methods that were used to collect the data. The data collected is considered meaningful due to a subfield’s approval of the materials and methods. This approval is the backing of claims made in every- day science (e.g. ‘paradigm’ science in the terms of [Kuhn2012]). With this model, truth-bearing Statements (possibly with Qualifiers) from biomedical communica- tions are explicitly connected to the Data, Material, and Methods. Attribution also has a place (for instance justifying use of a method by citing it in the literature). A subargument can be represented as a micropublication for a related claim, however far back we wish to extend the network.