Formalizing knowledge and evidence about potential drug drug interactions, BDM2I at ISWC2015,...
-
Upload
jodischneider -
Category
Technology
-
view
1.085 -
download
0
Transcript of Formalizing knowledge and evidence about potential drug drug interactions, BDM2I at ISWC2015,...
Formalizing knowledge and evidence about potential drug-drug interactionsJodi Schneider1, Mathias Brochhausen2, Samuel Rosko1, Paolo Ciccarese3,6, William R. Hogan4, Daniel Malone5, Yifan Ning1, Tim Clark6, Richard D. Boyce1
1 – University of Pittsburgh2 – University of Arkansas for Medical Sciences3 – PerkinElmer Innovation Lab4 – University of Florida5 – The University of Arizona6 – Massachusetts General Hospital and Harvard Medical School
Potential Drug-Drug Interaction (PDDI)
• Co-prescription or co-administration of two drugs known to interact.
• Preventable and have large impact:– 5-14% of hospital inpatients (Magro et al 2012)
– .02-.17% of emergency room visits each year (CDC)
2
There is substantial disagreement in PDDI evidence sources
• 4 clinically oriented drug information compendia agreed on only 2.2% of 406 PDDIs considered to be “major” by at least one source. (Abarca et al 2003)
• Only 1/4th of 59 contraindicated drug pairs were listed in 3 PDDI information sources. (Wang et al 2010)
• Only 18 (28%) of 64 pharmacy information and clinical decisions support systems correctly identified 13 PDDIs considered clinically significant by a team of drug interaction experts. (Saverno et al 2011)
3
Goal: Make disagreements and discrepancies visible
• Did compendium A and compendium B consider the same evidence?
• Cite a particular statement, indicating a disagreement with it.
• Mark some evidence as more or less credible.
4
Overall project goals
• Support drug information compendia editors who review PDDI evidence.
• More easily integrate and cross-check info.• Update and maintain an existing knowledge
base, the Drug Interaction Knowledge Base (DIKB).
5
Update an existing knowledge base
• The old knowledge base, DIKB-old, has:– PDDI claims & evidence collected in prior work– 36 competency questions– A taxonomy of PDDI evidence
• Limitations– Not used clinically– No computable, standard representation of PDDI
knowledge claims and evidence
6
Requirements
1. Distinguish between:– a drug drug interaction
(an actual occurrence in a patient)– a potential drug drug interaction
(an information content entity that may exist because of an observation or inference).
2. Link to biological processes3. Computability4. Maintainability
7
Main idea
• Formalize claims• Use new ontologies
– Nanopublications– Micropublications– Updated OBO Foundry ontologies, including a
bespoke ontology for Potential Drug-Drug Interactions (DIDEO)
8
Process
• Collaborate with an evidence workgroup– includes 4 pharmacology experts:
Carol Collins, Amy Grizzle, Lisa Hines, John R Horn• Investigate how to formalize PDDI claims and
evidence, focusing on clinically relevant info.• Transform data into the new DIKB.
– 410 knowledge claims about 88 drugs (October 2015)
9
USING NEW METHODS TO FORMALIZE PDDI CLAIMS AND EVIDENCE
1. Respect clinically-relevant distinctions
• Observed– PDDI evidence derived from an in vivo study
• Inferred – PDDI Claim inferred from metabolic mechanistic
knowledge of how drugs interact
11
2. Pull in external knowledge
• Compatible with OBO Foundry• Reuse identifiers:
– ChEBI - Chemical Entities of Biological Interest– PRO - PRotein Ontology
• Flexible, query-based infrastructure
12
3. Use nanopublications (NP)
13
assertion
provenance
publication info
nanopublication
Based on http://nanopub.org/wordpress/?page_id=65
assertion
prov:generatedAtTime prov:hadMember prov:wasAttributedTo prov:wasDerivedFrom
DIKB nanopublication
prov:generatedAtTime prov:wasAttributedTo
15
4. What about the EVIDENCE? DATA?
Kantola, T., Kivistö, K. T., & Neuvonen, P. J. (1998). Erythromycin and verapamil considerably increase serum simvastatin and simvastatin acid concentrations*. Clinical Pharmacology & Therapeutics, 64(2), 177-182.http://dx.doi.org/10.1016/S0009-9236(98)90151-5
Model the evidence with the Micropublications Ontology
A claim is supported by methods, materials, and data: – MP:Claim– MP:Method– MP:Materials– MP:Data
16
Clark, T., Ciccarese, P., Goble, C.: Micropublications: a semantic model for claims, evidence, arguments and annotations in biomedical communications. Journal of Biomedical Semantics 5(1), 28 (2014).
Ontology at http://purl.org/mp16
MP is complementary to NP
MP NP• Publishable• Citable• Use provenance
17
• Publishable• Citable• Use provenance
MP is complementary to NP
MP NP• Publishable• Citable• Use provenance• Logical structure
18
• Publishable• Citable• Use provenance• Natural language• Supports claim conflict.• Makes evidence structure
explicit.
We link MP and NP
• We extended the MP ontology to add two new properties:– MP:formalizedAs– MP:formalizes
• NP <formal representation>MP:formalizes
MP:Claim <natural language representation>
19
ddi:ddi-spl-annotation-np-assertion-314 a np:assertion, owl:Class ; rdfs:label "erythromycin - simvastatin potential drug-drug interaction" ; rdfs:subClassOf
[ a owl:Restriction ; owl:onProperty obo:IAO_0000136 ; owl:someValuesFrom [ a owl:Class ; owl:intersectionOf ( obo:DIDEO_00000012 [ a owl:Restriction ; owl:hasValue obo:CHEBI_48923 ; owl:onProperty obo:BFO_0000052 ] ) ] ],
[ a owl:Restriction ; owl:onProperty obo:IAO_0000136 ; owl:someValuesFrom [ a owl:Class ; owl:intersectionOf ( obo:DIDEO_00000013 [ a owl:Restriction ; owl:hasValue obo:CHEBI_9150 ; owl:onProperty obo:BFO_0000052 ] ) ] ],
obo:DIDEO_00000000 .
NP: Assertion
MP:Claim
Building up an MP graph
Building up an MP graph
Building up an MP graph
Building up an MP graph
Building up an MP graph
Building up an MP graph
Building up an MP graph
Building up an MP graph
Building up an MP graph
OUTLOOK
Future Work
• Can clinicians quickly identify the rationale for any given knowledge claim, using the DIKB?
• Are MP’s easy to create with appropriate tools?• Can drug experts retrieve more complete
information with search tools derived from the DIKB?
32
Thanks!
• Carol Collins, Amy Grizzle, Lisa Hines, Harry Hochheiser, and John R Horn
• ERCIM “Alain Bensoussan” Fellowship Programme (FP7/2007-2013) no 246016
• National Library of Medicine 1R01LM011838-01• National Library of Medicine & National
Institute of Dental and Craniofacial Research 5T15LM007059-29
33
Data is publicly available
• Nanopublication and Micropublication graphshttp://purl.org/net/drug-interaction-knowledge-base/published-NP-and-MP-graphs
• Querieshttp://purl.org/net/drug-interaction-knowledge-base/micropublication-queries
• DIDEO: Drug-drug Interaction and Drug-drug Interaction Evidence Ontology (Brochhausen et al 2014)
http://www.obofoundry.org/ontology/dideo.html34
Silos of PDDI knowledge claims
Scientific literatureProduct labeling Clinical experience
Evidence of PDDI knowledge claims is distributed across multiple sources
Pre-market studies Post-market studies
Product labeling
Reported in
Clinical experience
Scientific literature
Rarely reported in
Rarely reported inReported in
Rarely reported in
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
Source forSource for
Efficiency and scalability from structured publication
38
Facts in the evidence base are used for reasonable extrapolation…
39
Making an inference
40
Micropublication Ontology (MP)
Clark, T., Ciccarese, P., Goble, C.: Micropublications: a semantic model for claims, evidence, arguments and annotations in biomedical communications. Journal of Biomedical Semantics 5(1), 28 (2014).
Ontology at http://purl.org/mp 41
DIKB micropublication graph for the erythromycin - simvastatin
interaction
42
ddi:ddi-spl-annotation-np-assertion-314 a np:assertion, owl:Class ; rdfs:label "erythromycin - simvastatin potential drug-drug interaction" ; rdfs:subClassOf
[ a owl:Restriction ; owl:onProperty obo:IAO_0000136 ; owl:someValuesFrom [ a owl:Class ; owl:intersectionOf ( obo:DIDEO_00000012 [ a owl:Restriction ; owl:hasValue obo:CHEBI_48923 ; owl:onProperty obo:BFO_0000052 ] ) ] ],
[ a owl:Restriction ; owl:onProperty obo:IAO_0000136 ; owl:someValuesFrom [ a owl:Class ; owl:intersectionOf ( obo:DIDEO_00000013 [ a owl:Restriction ; owl:hasValue obo:CHEBI_9150 ; owl:onProperty obo:BFO_0000052 ] ) ] ],
obo:DIDEO_00000000 .
NP: Assertion
ddi:ddi-spl-annotation-np-assertion-314
prov:generatedAtTime "2015-08-13T15:46:38.746060"^^xsd:dateTime ; prov:hadMember dikbEvidence:EV_PK_DDI_NR, dikbEvidence:EV_PK_DDI_Par_Grps, dikbEvidence:EV_PK_DDI_RCT ;
prov:wasAttributedTo <http://orcid.org/0000-0002-2993-2085> ;
prov:wasDerivedFrom "Derived from the DIKB's evidence base using the listed belief criteria" .
NP:Provenance
ddi:ddi-spl-annotation-nanopub-314
prov:generatedAtTime "2015-08-13T15:46:38.745973"^^xsd:dateTime ;
prov:wasAttributedTo <http://orcid.org/0000-0002-2993-2085> .
NP:Publication Info
46
Scalability
47
48
4. What about the EVIDENCE?