Transcript The Drug Interaction Knowledge Base

Jodi Schneider, Samuel Rosko, Yifan Ning, Richard D. Boyce
Department of Biomedical Informatics, School of Medicine, University of Pittsburgh
Overview
Why?
• Knowledge is buried in a haystack of trials and papers.
• Need deeper information than title/abstract/authors/venue.
• Huge amount of human time and effort is spent in “keeping up”.
From Annotations to Knowledge Base
Source documents are annotated to extract relevant data:
Domain/
model of/
evidence
Source/
document
Domain/model/of/evidence
Source/document
Annotation
Annotation
Eight male and four female
healthy volunteers (age range,
20 to 29 years; weight range,
54 to 80 kg)
Structured publication
Materials
1
The Lab
500 mg erythromycin (Ery-Max, 250 mg tablet; Astra, Sodertalje, Sweden),
80 mg verapamil (Isoptin, 80 mg tablet; Knoll, Ludwigshafen, Germany)
Data
1
matched placebo
2,40 mg simvastatin
The TRanslational Informatics Applied to Drug Safety (TrIADS) Lab and
our collaborators are constructing a Drug Interaction Knowledge Base.
Aggregate
Knowledge/base
Source/
document
Evidence/base
Source/
document
Domain/
model of/
evidence
3
Domain/
model of/
evidence
Source/
document
Annotation
Domain/
model of/
evidence
2
Our work is part of “Addressing Gaps in Clinically Useful Evidence on
Drug-Drug Interactions”, an NLM R01 grant. Three groups contribute to
the knowledge base’s Meta-data Standard:
1. Qualitative Inquiry Workgroup
2. Standard Development Workgroup
3. Evidence Panel Workgroup
Annotation
Filter
Structured publication/meeting/
belief/criteria/set/A
Source/
document/
1
Annotation
Domain/
model of/
evidence
Annotation
Structured
publication/3
Annotations index and can be used to provide access to each source document.
Structured
publication/2
Structured
publication/1
Annotations are stored in the evidence base (EB). Filtering the EB generates a knowledge base.
Micropublications
In the evidence base, we model knowledge claims about drug-drug
interactions and the supporting evidence for these claims.
obo:CHEBI_48923
obo:CHEBI_9150
mp:qualifiedBy
The Drug Interaction Knowledge Base
MP
1
The Drug Interaction Knowledge Base contains:
mp:argues
mp:qualifiedBy
Claim
1
obo:DIDEO_00000000
mp:qualifiedBy
erythromycin increases the AUC of simvastatin
mp:supports
73 drugs
Data
1
354 claims of potential drug-drug interactions in the evidence
base
94 claims of potential drug-drug interactions in the knowledge
base
mp:supports
mp:supports
Clinical)experience
Post$market)studies
Pre$market)studies
Rarely(reported(in
Reported(in
Reported(in
Materials
1
Using the Micropublication Ontology, we model published claims as formal assertions linked to
primary data and resources. Claims may be mutually inconsistent.
Each Micropublication links a claim to its supporting evidence and source document.
Potential drug-drug interactions are a significant source of preventable
drug-related harm. Unfortunately, most drug information sources
disagree substantially in their content. One contributing factor is that
there is no standard way to represent PDDI knowledge claims and
associated evidence in a computable form.
http://dx.doi.org/
10.1016/
S0009-9236(98)90151-5
Method
1
Contents of the Drug Interaction Knowledge Base, as of August 2015:
http://purl.org/net/drug-interaction-knowledge-base/published-NP-and-MP-graphs
Why Drug-Drug Interactions?
mp:supports
Our Approach
The underlying strategy is to
(1) construct both an evidence base and a knowledge base
(2) model knowledge with ontologies; and
(3) annotate the scientific literature and other source documents.
We use belief criteria to determine what sort of evidence is required
to assert that a claim is ‘True’. We generate a knowledge base from
the evidence base. For each claim meeting the belief criteria, we
publish a Nanopublication.
Rarely(reported(in
Scientific)literature
Rarely(reported(in
Product)labeling
Source(for
Source(for
Drug)Compendia))synthesize)PDDI)evidence)into)
knowledge)claims but
• May)fail)to)include)important) evidence
• Disagree)if)specific)evidence)items)can)support)
or)refute)PDDI)knowledge)claims
Evidence relevant for establishing PDDI knowledge claims is distributed across product
labeling, the scientific literature, case reports, and other sources.
University of Pittsburgh
We distinguish the evidence base from the knowledge base. This helps clarify disagreements.
Nanopublications indicate the belief criteria under which they are ‘True’.
Funded by training grant 5T15LM007059-29 from the National Library of Medicine/National Institute of Dental and Craniofacial Research and by 1R01LM011838-01 from the
National Library of Medicine. External collaborators include Tim Clark (MGH/Harvard), Mathias Brochhausen (University of Arkansas for Medical Sciences).
Department of Biomedical Informatics