Linking Open Drug Data
Susie Stephens,
Principal Research Scientist, Eli Lilly
The Linked Data Cloud
Source: Chris Bizer
Linking Open Drug Data
• HCLSIG task started October 1, 2008
• Primary Objectives
• Survey publicly available data sets about drugs
• Publish and interlink these data sets on the Web
• Explore interesting questions in competitive
intelligence that could be answered if the data sets
are linked
• Participants: Bosse Andersson, Chris Bizer, Kei Cheung, Don
Doherty, Oktie Hassanzadeh, Anja Jentzsch, Scott Marshall, Eric
Prud’hommeaux, Matthias Samwald, Susie Stephens, Jun Zhao
Assessment of Data Sources
Mark Sharp et al. A Framework for Characterizing Drug Information Sources. AMIA 2008
Published Data Sets
• LinkedCT (http://linkedct.org)
• Online registry of more than 60,000 clinical trials
• Published in XML
• 7,011,000 triples (290,000 interlinking)
• DrugBank (http://www4.wiwiss.fu-berlin.de/drugbank)
• A repository of almost 5,000 FDA-approved drugs
• Published as DrugBank DrugCards
• 1,153,000 triples (23,000 interlinking)
• DailyMed (http://www4.wiwiss.fu-berlin.de/dailymed/)
• High quality information about marketed drugs
• Flat file representation
• 124,000 triples (29,600 interlinking)
• Diseasome (http://www4.wiwiss.fu-berlin.de/diseasome)
• Information about 4,300 disorders and disease genes linked by known disorder-gene
associations
• Published in XML
• 88,000 triples (23,000 interlinking)
Classes of Links
• Based on common identifiers
• Links present in the source data sets
• Based on link discovery and record linkage techniques
• String matching
– E.g., “Alzheimer’s disease” in LinkedCT was matched with
“Alzheimer_disease” in Diseasome
• Semantic matching
– E.g. “Varenicline” has the synonym “Varenicline Tartrate” and the brand
names “Champix” and “Chantix”
Business Use Case
• A neuroscience focused business manager is interested in seeing an
update on new clinical trials by competitors on Alzheimer’s Disease (AD)
• A phase III trial by Pfizer for a drug called Varenicline has just been listed in
linkedCT
• More information of interest is found in DBpedia, DailyMed, and DrugBank
• DailyMed indicates the drug is already on the market for Nicotine addiction
and has minimal side effects
• DrugBank allows the manager to see the targets for Varenicline
• Diseasome, however, indicates that the corresponding genes are only
implicated in nicotine addiction, rather than AD
• This suggests a more complex relationship between the diseases than just
the drug target
• Extending the browsing to the SWAN Knowledgebase shows that there are
hypotheses relating AD to nicotine receptors through amyloid beta
Technical Challenges
• Life sciences data is difficult to connect due to inconsistent
terminology and the prevalence of synonyms, and homonyms
• Refinement of tools and techniques for enabling more automatic
linking of entities across data sets
• Selection of ontologies to enable consistent mappings
• Development a sufficiently robust platform as to enable
inferencing
• Provide an interface to users that supports browsing, querying,
and filtering data
• Persuade data providers to publish in RDF would alleviate the
need for us to update data, and provide some of the interlinking
Next Steps
• Ensure that existing data are accurately and comprehensively
linked
• Incorporate additional data sources into the LODD cloud that
are of interest to competitive intelligence (e.g. Traditional Chinese
Medicine)
• Use novel link discovery tools and frameworks including Silk
and LinQuer
• Explore using SIOC to aggregate information as what patients
are saying about drugs
• Submit paper to the iTriplify Challenge
Task Alignment
• LODD is looking to use Pharma Ontology’s work to help
inform the mappings
• Data converted to RDF is also loaded into BioRDF’s
HCLS KB
Conclusions
• Added 4 drug-related data sets into the cloud for
competitive intelligence
• Will add further data sources to the LODD cloud to
enable more insights to be gleaned
• Will continue to explore and test tools that are being
developed for LOD