Providing RDF Metadata for Biological
Databases using Semantic Overlays
Ben Szekely
Lee Feigenbaum (not present but highly responsible)
May, 2006
© 2006 IBM Corporation
IBM Internet Technology
LSID Overview - Syntax
 5 Part Format: urn:lsid:authority:namespace:object[:revision]
–
–
–
–
–
urn:lsid
– Mandatory prefix
authority
– Unique string, e.g. domain name of organization
namespace
– Alphanumeric sequence that constrains the scope
– E.g. to a particular database, species, etc…
object
– Alphanumeric sequence describing the object
[revision]
– Optional alphanumeric sequence describing the version of the object
 Example: urn:lsid:ncbi.nlm.nih.gov:genbank:af271072
 Example: urn:lsid:pdb.org:pdb:1aft:1
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
LSID Overview - Resolution
1a - DDDS NAPTR
1b - SRV Record Lookup
DNS/DDDS
2 - getAvailableServices()
LSID Authority
Client
WSDL
3a - getData()
3b - getMetadata()
Data Service
Metadata Service
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
LSID Overview – Comparison with URLs
URL
LSID
 Tied to physical addresses
 Server structure may change frequently
 Brittle (broken links)
 One location only
 One protocol per URI
 Same name = same content, always
 Location independent
 Enables transparent caching
 Formalized, rich multi-sourced metadata
retrieval
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
The RDF Roadblock
 Consider the great corpus of biological data that could (should?) exist in
RDF.
–
NCBI, Ensembl, GeneOntology, HUGO, PDB
–
Metadata linking experiments, specimens, research papers
 Infeasible to convert all of this data to RDF
–
Development cost
–
– Ontology development, data conversion, query optimization
Standards issues
–
– When data is physically stored as RDF, ontologies are quite important to get right initially.
– Standards bodies have spent a long time settling on XML-based standards over the past
10 years, do they really want to do it all over again?
Lack of technology and tooling
– Does the technology really exist to host the NCBI online database and Web Service on a
system backed entirely by RDF?
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
Semantic Overlays – a great start
 Provide an RDF-view of a database
–
Expose logical subsets of the database as named graphs
– Ex: a Pubmed article, an NCBI Protein
–
Named graphs should be exposed as URLs or LSIDs
– Standard resolution
– Simple to include in SPARQL queries
–
Ontologies in the view can evolve over time
– No need to modify the body of data when the ontology changes
– Perfect for experimentation, community feedback
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
Semantic Overlays – implementation techniques
 XML/Web Services
–
XSLT: XML -> RDF/XML (or N3)
–
XML-Beans to RDF-Beans (i.e. Castor to Jastor)
 HTML Pages
–
Screen scraping -> RDF
 Relational databases
–
SQL Result Sets -> RDF (Jena, Jastor)
Many other techniques are possible
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
Semantic Overlay - limitations
 Cannot query the complete database with SPARQL
– (We can, however, provide a SPARQL extension that wraps the
native query interface of the database. Ex. SPARQL Entrez
Search)
 RDF Translation can be expensive
 RDF Linkage is limited by the intrinsic linkages in
the store Ex. (Genbank -> Pubmed)
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
Semantic Overlay – Demo 1
 Provide a simple, one-stop answer to the question:
How can I discover proteins that are
relevant to my work and locate
antibodies that target those proteins?
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
SPARQL is…
 …a query language for selecting values
from RDF graphs
 …a protocol for issuing queries via
HTTP GET, HTTP POST, or SOAP
 …a W3C Candidate Recommendation
 …capable of returning results serialized
as web-friendly JSON structures
 …perfect for mashing up disparate data
sources representable as RDF
PREFIX foaf: <…foaf/0.1/>
PREFIX rdf: <…22-rdf-syntax-ns#>
SELECT ?name ?email
WHERE {
?person rdf:type foaf:Person .
?person foaf:name ?name .
OPTIONAL {
?person foaf:mbox ?email .
}
}
?name
?email
Lee Feigenbaum
feigenbl@us.ibm.com
Grandma Feigenbaum
(unbound)
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
The Data Sources
 Entrez protein sequence and gene databases
–
National Center for Biotechnology Information (NCBI)
–
http://www.ncbi.nlm.nih.gov/
–
RDF  LSID metadata
 Antibody directory
–
Alzheimer Research Forum (AlzForum)
–
http://alzforum.org/res/com/ant/default.asp
–
RDF  HTML scraping
 Mapping data between genes and antibodies
–
Alan Ruttenberg, Millennium
–
RDF  spreadsheet data
 Taxonomy information
–
Wikispecies, free species directory
–
http://www.wikispecies.org
–
RDF  XSLT applied to XHTML
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
The Tools
 JavaScript SPARQL client library
–
Issue SPARQL SELECT queries and retrieve results as JavaScript objects
–
Supports all SPARQL endpoints returning JSON results (SPARQLer, Rasqal,
XMLArmyKnife, …)
–
http://www.thefigtrees.net/lee/sw/sparql.js
 JSON
–
Lightweight serialization of data structures (e.g. SPARQL resultsets)
–
http://www.json.org
 Microtemplates
–
Automagically bind JavaScript-object data to DHTML fragments
–
http://www.microtemplates.org
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
The Demo
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
What We Learned
Take-away Lessons
 “With a query language, a client
can design their own interface.”
- Leigh Dodds
 SPARQL + JSON is a powerful
Web 2.0 environment
 Even data sources not natively
expressed in RDF can be mashed
up with SPARQL
 Life sciences provides a rich
domain of situational problems to
approach with SPARQL-based
mashups
Looking Ahead
 As we deal in larger and larger
data sets, on-the-fly RDF creation
becomes impractical, so:
–
“Smart” federation
–
Dedicated SPARQL endpoints
 Universal naming, merged graphs,
and shared predicates only get us
so far, so:
–
Custom relations
–
owl:sameAs
–
Human-guided curation
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
Semantic Overlays – Demo 2
 Annotating Semantic Overlay Named Graphs using LSID
Goal: add custom, resolvable metadata to NCBI objects.
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
Demo Technologies
 Insightlink Annotation System
–
Annotations stored in RDF/DB2
–
Custom annotation forms based on identity, context, application
–
Plugins for I.E., PowerPoint, Word, Acrobat, Spotfire
 LSID/NCBI Proxy
–
Web proxy to add LSID tags to NCBI search results
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation
IBM Internet Technology
Thanks!
 Questions?
 More information: bhszekel@us.ibm.com
 Demo online at http://thefigtrees.net/lee/sw/demos/antibodies/
 Thanks to:
–
Alan Ruttenberg, Millennium
–
June Kinoshita and Colin Knep, Alzheimer Research Forum
–
Lee Feigenbaum, Elias Torres, Stephen Evanchik, and Alister Lewis-Bowen, IBM
Providing RDF Metadata for Biological Databases using LSID-based Semantic Overlays
© 2006 IBM Corporation