Bio2RDF: Towards a mashup to build bioinformatics knowledge systems
Francßois Belleau a,*, Marc-Alexandre Nolin a,b,*, Nicole Tourigny b
, Philippe Rigault a
, Jean Morissette a,c
a
Centre de Recherche du CHUL, Université Laval, 2705 Boulevard Laurier, Que., Canada G1V 4G2
b
Département d’informatique et de génie logiciel, Université Laval, Cité Universitaire, Que., Canada G1K 7P4
c
Département d’anatomie-physiologie, Université Laval, Cité Universitaire, Que., Canada GIK 7P4
a r t i c l e i n f o
Article history:
Received 1 September 2007
Available online 21 March 2008
Keywords:
Knowledge integration
Bioinformatics database
Semantic web
Mashup
Ontology
a b s t r a c t
Presently, there are numerous bioinformatics databases available on different websites. Although RDF
was proposed as a standard format for the web, these databases are still available in various formats.
With the increasing popularity of the semantic web technologies and the ever growing number of databases
in bioinformatics, there is a pressing need to develop mashup systems to help the process of bioinformatics
knowledge integration. Bio2RDF is such a system, built from rdfizer programs written in JSP,
the Sesame open source triplestore technology and an OWL ontology. With Bio2RDF, documents from
public bioinformatics databases such as Kegg, PDB, MGI, HGNC and several of NCBI’s databases can
now be made available in RDF format through a unique URL in the form of http://bio2rdf.org/namespace:id.
The Bio2RDF project has successfully applied the semantic web technology to publicly available
databases by creating a knowledge space of RDF documents linked together with normalized URIs and
sharing a common ontology. Bio2RDF is based on a three-step approach to build mashups of bioinformatics
data. The present article details this new approach and illustrates the building of a mashup used to
explore the implication of four transcription factor genes in Parkinson’s disease. The Bio2RDF repository
can be queried at http://bio2rdf.org.
Ó 2008 Elsevier Inc. All rights reserved.
1. Introduction
A rapid way to look for information on the web is to use a search
engine such as Google. The results, however, are a list of suggested
HTML pages devoid of context and semantics and requiring human
interpretation. For a more contextual search in the field of molecular
biology, a specialized tool like NCBI’s Entrez [1] is more effective
because it is dedicated to the specific domain under
consideration. The Entrez search engine uses all the different databases
hosted by NCBI; its data integration approach, based on
hyperlinks, is illustrated by its database schema (http://
www.ncbi.nlm.nih.gov/Database/). The Kegg’s DBGET [2] search
service is another example of a specialized search engine dedicated
to genes and pathways.
Each year, NAR [3] publishes a new version of its bioinformatics
database list. In the 2006 issue, over one thousand servers were
listed. Other specialized lists of databases are now available. For instance,
the Pathguide website [4] lists 244 pathways and protein
interaction databases. With such a proliferation of knowledge
sources, there is a pressing need for a global multisite search
engine and for good data integration tools. According to the data
warehouse approach, such services can be built by collecting information
into a central data repository [5] and queried with an interface
built on top of the repository. However, the warehouse
approach does not address the problem of accessing a database
outside the warehouse. A system that would be able to query
and connect different databases available on Internet would solve
that problem. This is one of the goals of the semantic web approach:
to offer the data warehouse experience without the need
of moving first the data into a central repository.
To address the data integration problem, the semantic web
community, led by the W3C, proposed a solution based on a series
of standards: the RDF format for document [6] and the OWL language
for ontology specification [7]. RDF and OWL generate a series
of entities called ’triple’ in the form of a subject, predicate and object.
Database systems able to handle triples are called triplestore.
New software has been created by the computer science community
to exploit them. Some tools are still in the development stage,
others are mature enough to be used in production systems, like
the open source project Sesame [8], which is a triplestore server
providing storage and querying capabilities.
We have developed a semantic web application called Bio2RDF
to help solve the problem of knowledge integration in bioinformatics.
Bio2RDF uses RDF documents and a list of rules to create URIs
1532-0464/$ - see front matter Ó 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.jbi.2008.03.004
* Corresponding authors. Address: Département d’informatique et de génie
logiciel, Université Laval, Cité Universitaire, Que., Canada G1K 7P4. Fax: +1 418
525 4444x42761 (M.-A. Nolin).
E-mail addresses: francoisbelleau@yahoo.ca (F. Belleau), Marc-Alexandre.Nolin@genome.ulaval.ca,
lotus@ieee.org (M.-A. Nolin).
Journal of Biomedical Informatics 41 (2008) 706–716
Contents lists available at ScienceDirect
Journal of Biomedical Informatics
journal homepage: www.elsevier.com/locate/yjbin
that will create linked data. Bio2RDF can be seen as a mashup
application because it combines data from more than one source,
following the definition of a mashup given in Wikipedia [9]. Indeed,
Bio2RDF integrates publicly available data from some of
the most popular databases in bioinformatics. As a mashup is more
often associated with a graphical user interface than data (or
knowledge) integration, Bio2RDF can be described as a data mashup
using a semantic web approach for data (or knowledge) integration.
The purpose of the present paper is to describe the data
integration approach used with Bio2RDF.
1.1. Integration methods in bioinformatics
The idea of integrating data from various sources is not a recent
concern in bioinformatics, as illustrated by the research work of
Davidson [10], Köhler [11], and Stein [5].
In 1995, Davidson [10] suggested the following basic steps to
integrate bioinformatics data: transformation to a common data
model, matching of semantically related objects, schema integration,
transformation of data into a federated database, and finally
matching of semantically equivalent data. Davidson et al. suggested
to ‘‘Transform data to the federated database on demand”.
This solution can now be achieved in a semantic web approach
through the Bio2RDF project, where data is transformed into RDF
format.
In 2003, the Semeda (Semantic Meta Database) [11] was another
attempt at integrating heterogeneous databases. Kohler
identified four problems. (1) In different databases the same things
can be given different names. This is the case with the two pathway
databases, Kegg [12] and Reactome [13]: they both annotate
and describe the same pathways in completely different semantic
spaces. (2) Attribute names are not self-explanatory. For example
the way of specifying URLs should always be the same, as in the
HTML href attribute. (3) Querying databases requires knowledge
about its contents. This is exactly what the semantic web approach
wants to avoid. (4) Due to the lack of a systematic linking mechanism,
only the most important attributes are associated. Therefore,
a normalization of identifiers is mandatory. Such a normalization
was the goal of the LSID [14] project.
Also in 2003, Stein [5] highlighted three approaches typically
used by data integrators: link integration, view integration, and
data warehousing. The first one uses the linking capability of the
web; the second one is the creation of portals that aggregate the
information; the third, data warehousing, stores everything in a
single unified database. Stein also proposed an ontological approach
that he called knuckles-and-nodes. Simply stated, this
approach is about building databases of links between data, but
not storing any of it. This strategy is very similar to that of Bio2RDF.
1.2. Integration using a semantic approach
Ontology design is not a new topic in bioinformatics, however
projects using the OWL language are new. Tambis [15], BioPAX
[16] and UniProt [17] are three projects which have adopted this
new formalism. Describing and building knowledge systems using
the semantic web’s RDF standard as a knowledge representation
format is still a challenge and several projects such as YeastHub
[18] and FungalWeb [19] have explored this research topic.
In 2000, TAMBIS [15] was the first project to propose a unified
ontology described in OWL and covering many aspects of the bioinformatics
knowledge space. The BioPAX ontology [16], a more recent
proposition with the same goal, is already used by six
pathway database websites. The UniProt consortium has made
available an RDF version of the UniProt protein knowledge base
through their new beta website (http://beta.uniprot.org). The documented
translation [20], describing the migration from the UniProt
traditional text format to an RDF document has been a
guideline for the Bio2RDF project. Its ontology [21], available in
OWL format, was created with the Protégé ontology editor [22].
The YeastHub [18] project was the first attempt to build an integrated
database in RDF format unified by the Sesame’s triplestore.
The resulting warehouse of yeast genome data illustrates the potential
of the query capabilities afforded by a knowledge base once
the document’s URIs have been normalized. The Bio2RDF approach
is similar to that of YeastHub, with the exception that Bio2RDF is
open source, extensible and provides access to millions of documents
from hundreds of different organisms.
The FungalWeb [19] project also focused on data integration,
specifically for the needs of industrial enzyme biotechnology. An
instantiated OWL-DL ontology was designed using Protégé and
the graphical query composer OntoIQ [23], in conjunction with Racer
and its query language nRQL. The interrogation of the integrated
knowledge base was illustrated by using application
scenarios. Instead of using Sesame, this research project employed
the commercial OWL reasoner Racer [24] which offers inference
capabilities.
A third integration project using RDF, conducted by Stephens
[25], integrated disparate biomedical data sources to help the drug
discovery effort. Different data sources were merged together: UniProt,
OMIM [26], Entrez Gene [27], Kegg, Gene Ontology [28], Intact,
Affymetrix probesets annotations and some others. This list
of major data sources is similar to that of Bio2RDF. To build this
knowledge base system, Stephens used the Oracle RDF data model
as the triplestore and the Seamarks Navigator for faceted browsing.
Bio2RDF is also an integration project making bioinformatics data
available on the web from various data sources, but uses open
source software. This framework does not offer a user interface
with faceted browsing, but tools like Simile Exhibit can be used directly
with the Bio2RDF data.
In a review about data integration and genomic medicine [29],
the authors have identified two axes defining the integration approach.
The first one describes the architecture of the system, the
second axis defines the knowledge description. Using this definition,
Bio2RDF should be classified into Peer data management systems
with an ontology knowledge description.
Several lessons were learned from these experiences. Firstly, the
semantic web approach can be used effectively to integrate bioinformatics
data. Secondly, knowledge bases created thus far were
designed to answer specific questions. Thirdly, if one wants to promote
the semantic web method for data integration, the use of free
open source software should be encouraged in order to enhance
the reproducibility of results that are published in the literature.
The Bio2RDF project was built as a result of these lessons.
The present article intends to show how Bio2RDF merges bioinformatics
knowledge from different sources. Aggregation of related
knowledge sources should eventually be as easy as dragging and
dropping them into a knowledge store. The Bio2RDF integration
technology is built on programs found in the open source community:
the Sesame triplestore and Elmo RDF crawler [8], JSP and JSTL
[30] which are technologies used to generate web pages and the
URLrewrite library [31] used to proxy HTTP requests. RDF-formatted
documents, required by semantic web technologies, are not yet
common on Internet. At this time, only UniProt and GO websites
offer RDF documents to build semantic web applications. One of
the main goals of the Bio2RDF project is to convert into RDF format
documents available from public databases. Bio2RDF is a flexible
open source software which allows to develop new rdfizer programs
in order to add new knowledge sources or experimental private
data. The result section below shows, through a use case, how
the Bio2RDF mashup system can be used to build a triplestore that
supports the exploration of the Parkinson’s disease knowledge
space.
F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716 707
2. Materials and methods
Two main ideas have oriented our software development: the
conversion of existing databases into RDF format (a process called
‘‘rdfizing”) and the use of existing semantic web software to merge,
query and visualize the data. These software components are: the
Sesame open source triplestore, the Protégé ontology editor, the
Piggy Bank [32] semantic web browser plug-in for FireFox and
the Welkin [33] RDF graph visualizer, both developed by the MIT,
and finally the experimental LSID browser [34].
We first show the method used to build the ontology. We then
explain how to use rdfizer programs to transform existing documents
into RDF format and how we normalized URIs. This section
ends with a high level description of the system software
architecture.
2.1. Ontology design
An ontology can be defined as an explicit specification of a conceptualization,
a conceptualization being an abstract and simplified
view of the world that needs to be represented for some
purpose [35,13]. For a given knowledge base or knowledge system,
it means that a conceptual language should be used to define the
objects and the relations to be represented. OWL is the conceptual
language chosen by the semantic web community for ontologybased
knowledge representation. To design the ontology of
Bio2RDF, we used the Protégé open source framework and its
OWL editor Protégé-OWL.
Since the main goal of Bio2RDF was to convert into RDF format
the documents available on the web (Entrez Gene description of
Hk1 from NCBI website for instance), the first step was to analyze
the existing HTML page to identify the predicates and relations
describing the entities. The label of a field corresponds to its predicate,
and the hyperlink corresponds to the URI of the resource,
usually defined in another namespace like GI, GO or PubMed. Using
this approach we produced an OWL description from each selected
HTML document. This step was repeated for each namespace recognized
by the current version of Bio2RDF: GO, OMIM, PDB, etc.
For BioPAX and UniProt, this step was unnecessary because their
OWL schema was already available. Finally, the global bio2rdf-
2007-02.owl [36] ontology description was built by merging the
ontology file of each namespace.
After the Bio2RDF ontology was created, the second step consisted
in writing the necessary rdfizer programs in JSP in order to
address two key objectives: (1) mapping between the data elements
of the original document and the predicates in the RDF version,
(2) normalization of URI resources according to the Bio2RDF
syntax. The creation of rdfizer programs, performed for more than
twenty different namespaces, was the main task of the Bio2RDF
project.
The design of the Bio2RDF’s ontology was inspired by already
existing ontologies. For instance, rdf:type and rdfs:label were
systematically used in each document. The label predicate always
contains the name of the resource followed by a short
form of its URI enclosed with ‘‘[ ]”. For example, rdfs:label of
geneid:15275 is ‘‘hexokinase 1 (Hk1) [geneid:15275]”. Some
common predicates from the Dublin Core project [37] were used,
in particular dc:title, dc:identifier, dc:created and dc:modified.
We also used the FOAF [38] namespace to describe people and
the bibTeX [39] namespace for literature references. We had to
create our own predicates in the bio2rdf namespace, the most
frequently used ones being bio2rdf:url, bio2rdf:urlImage,
bio2rdf:xRef, bio2rdf:name and bio2rdf:synonym. The definition
of the semantics of these predicates can be found in the Bio2RDF
ontology file [36].
2.2. Rdfizer programs
In an ideal world, all the data would be available in RDF format
with complete normalization of URIs, and all documents on
Internet would consequently connect together automatically. But
this is not the case in the real world. At this time, what exists is
an HTML version of the data accessible through web pages. The
Bio2RDF project provides RDF formatted documents from several
data sources in a normalized way. A JSP toolbox has been created
to generate RDF files from locally stored databases or directly
from HTML documents accessed via http requests. JSP
tools were used to create rdfizers, which are programs transforming
existing data into an RDF representation [40]. Several
different sources of data can be rdfized: relational databases,
text files, XML documents, and HTML pages. For each type of
knowledge source, a JSP program converts the data from the original
source into the RDF format. These programs use XPath, regular
expressions or SQL queries to extract knowledge from the
original data. For example:
XML to RDF conversion with an OMIM record from NCBI: The
program ncbi-omim2rdf.jsp converts the XML representation of
OMIM records provided by the NCBI efetch service [41] into
RDF. First, this program receives in parameter the OMIM id for
the disease under consideration. In the beginning of the program,
it fetches the information from the NCBI website and
places the XML document in memory where the translation
work can be done. Second, the information for the document is
extracted according to the ontology previously created. In JSP,
the JSTL XML library can be used to navigate across the document
retrieved using efetch.
SQL to RDF conversion with an Ensembl record: Ensembl [42] provides
online access to its MySQL relational databases. The program
ensembl-g2rdf.jsp is used to do the conversion. The JSTL SQL library
offers functionalities to work with databases. The first action
is to establish a connection to the database using the parameters
that the providers (Ensembl in this example) supply on their website.
Once the connection is established, queries are created to
fetch all the data required to create the corresponding RDF
document.
Text file to RDF conversion with a Prosite record: Prosite [43] website
returns a text format description of a protein family domain.
The rdfizer prosite2rdf.jsp retrieves this text document, then uses
regular expressions to parse its content and to generate an RDF
version out of it.
The format of the RDF document produced by the Bio2RDF rdfizer
is not considered definitive. For this reason, the source code of
all our rdfizer programs has been made publicly available for
customization.
Although rdfizers on the Bio2RDF.org server or client can be
queried directly, they also have a REST-like [44] interface granted
by the use of an urlrewrite filter. This allows changing the underlying
programs without modifying the query methods, thus providing
stable URIs that can deal with changes in URIs by
upstream data providers. Such stable URIs are critical for properly
linked data and this subject is further elaborated in the following
section.
Bio2RDF is a three-step approach elaborated and tested for a
mashup of bioinformatics data. The first step is to build a list of
namespaces for different data providers. This enables the construction
of normalized URIs. The second step is to analyze a data source
to represent it in the RDF model. The third step is to build an rdfizer
that converts the information from the data source into its RDF
representation. The resulting RDF documents can then be put into
a triplestore in order to connect together. Further analysis can be
made on the triplestore with SeRQL or ‘‘REST like” queries.
708 F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716
2.3. URI normalization
The availability of RDF documents is not by itself sufficient to
obtain a mashup. External references, expressed as URIs, need to
be normalized to allow proper connection of triples. For example,
a PubMed reference with an identifier 12728276 can be referenced
with: PMID:12728276, pubmed:12728276 or PubMed:12728276.
For a knowledge agent, normalized representation of URI is mandatory
to ensure a functional connection between triples. Even in
existing well-formed RDF documents there is a problem with URIs.
For example the GO term of Hexokinase (GO:0004396) is referenced
by different URIs used by existing RDF data providers: UniProt,
OBO and BioPathways Consortium.
http://www.geneontology.org/go#GO:0004396
http://purl.uniprot.org/go/0004396
urn:lsid:geneontology.org.lsid.biopathways.org:go:0004396
Those are all supposed to represent the same concept: the definition
of Hexokinase molecular function according to Gene Ontology.
If we were to load these documents and use them in the same
triplestore, no links would be created around the Hexokinase concept
because the URIs are all different even if they correspond to
the same concept. By adopting the same URI pattern for all URIs
regardless the provider, the Bio2RDF system guarantees that the
connections are built automatically around the same concept.
The Bio2RDF URI synonym for the preceding published URI is:
http://bio2rdf.org/go:0004396
The Bio2RDF’s global strategy to ensure that the RDF graph refers
to unique concepts is accomplished by applying the Bio2RDF
URI syntax wherever possible. When an URI has already been assigned
to a graph by the data provider, we keep track of it by adding
an owl:sameAs predicate linking to this official URI.
Many proposals revolve around this subject, each and every one
having its pros and cons. The LSID proposal [14] is an identification
scheme using SOAP for content negotiation. The scheme makes it
possible to keep a stable URI even if the provider disappears because
it is domain independent, but at the cost of not being a routable identifier.
Another scheme is content negotiation with 303 redirections,
which would give routable URIs, but this behavior is not the default
one for a web server and the client has to be built to ask for XML/RDF
content type or else they will receive the HTML page instead of the
RDF one. The Bio2RDF project has established a simple set of rules
that data providers can apply to create URIs for their information:
1. Use a REST like interface. REpresentational State Transfer (REST)
enables us to produce a clear and stable URI for every document.
A default action can also be used, but it must be
explained on the data provider’s website. Either way, the data
provider should create a web page explaining her REST interface.
Also, a REST like interface does not need content negotiation,
web application or server redirection.
2. Lowercase all the URI up to the colon. The URI case sensitivity
poses a problem because each different case results in a theoretically
different URI. The two most successful kinds of URI
are domain names and email identifiers, which are case insensitive.
This allows addresses differing solely by case (such as
uniprot.org and UniProt.org) to refer to the same site. We suggest
converting into lower case the URI part up to the colon,
rendering it effectively case insensitive.
3. All URIs should return an RDF document. If an URI for a document
returns a web page instead of an RDF document, it is not easy to
connect its information directly with other linked data. Transformation
will need to be done before the RDF graph is
obtained. This very important rule is equivalent to rule #2 of
the Linked data design rules from the W3C [45]: ‘‘Use HTTP URIs
so that people can look up those names.” According to this rule,
using a HTTP dereferencable URI to identify a resource is natural
and very simple to use because the URI returns (dereference) an
RDF graph. Usage of LSID does not respect this specific rule
because there is no reference to any protocol in its URI.
Rules to convert URIs have been adopted for this reason. The
major one is that a URI is uniquely attributed to a document which
describes an object like an ontology term, a gene description or a
protein annotation. The syntax of a normalized URI is described
by the following pattern:
http://bio2rdf.org/<namespace>:<identifier>
For example, the identification for the article 12728276 from
PubMed would be written as http://bio2rdf.org/pubmed:12728276.
Our obvious URI design rule states that the document
URI corresponds to the unique URL which returns the
document in RDF format from Bio2RDF.org server. Consequently,
when a new document URI is added into the triplestore, the triples
referring to this document connect to it.
2.4. Bio2RDF architecture
Fig. 1 shows a schematic description of the Bio2RDF architecture.
All external data sources, in different formats (XML, Text,
ASN.1, KGML and RDF), are listed on the left part. These sources
are processed in two different ways. Websites, from which the
entire database was downloaded (MGI [46], HGNC [47], Kegg
[12], Entrez Gene [27], OMIM [26], GO [28], OBO [48], PDB
[49] and ChEBI [50]) to the Bio2RDF.org server, form the top
group. RDF documents from these sources are then accessible
at high speed since they are obtained directly from the Bio2RDF.org
server. Data from these websites is stored in a MySQL database.
Only requested documents from websites in the bottom
group (UniProt, Reactome, Prosite, PubMed, GenBank, PubChem,
etc.) are rdfized directly from the original source. In fact, the local
rdfizer program, part of the myBio2RDF application, queries
the data provider, transforms the returned document into normalized
RDF, and finally makes it available to the application.
The data from Entrez Gene, OMIM, OBO and Kegg, is cached
for availability and speed purpose: availability because few data
providers have an RDF version of their documents, and speed because
some data providers have restrictions for the access of
documents.
The new http://beta.uniprot.org server now offers a RDF graph
accessible by dereferencing HTTP URI of the form:
http://purl.uniprot.org/DATABASE/ID.rdf
Here DATABASE can be any of UniProt Consortium main databases:
uniprot, uriref, uniparc, etc. With millions of RDF documents
available from this single source at high speed, the only
transformation done by the Bio2RDF service consists in replacing
the syntax of URI. Ultimately, the data should always come live
from the data providers as done now with this new UniProt RDF
service. Then the Bio2RDF server would only act as a proxy server
forwarding the RDF query from the client requesting a graph to the
real data provider.
The myBio2RDF application contains two servlets running under
a Tomcat server: Elmo and Sesame. Elmo is a RDF crawler
which was originally created to follow rdfs:seeAlso predicate inF.
Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716 709
cluded in FOAF files. The Elmo capacity to crawl RDF documents
from the Bio2RDF.org website is applied to instantiate triples into
a local Sesame repository where the requested documents are
gathered. Next, the Sesame interface allows users to browse and
query the knowledge base with SeRQL. The Sesame version distributed
with the myBio2RDF package was slightly modified to fit
three special needs: (1) to allow its Explorer page to navigate
through the external link defined with bio2rdf:url; (2) to see
images defined by bio2rdf:urlImage; (3) to export query results
in tabular format compatible with spreadsheets. Three specific services
were added to allow Elmo to crawl specific knowledge:
(1) To obtain a list of URIs corresponding to the results of a text
search using the search engine of the corresponding website.
This tool is very useful because it leverages the existing text
search capability available from the official data provider.
http://localhost:8080/search:TEXT@database
where database = [omimjgeneidjpubmedjmeshjkeggjuniprot]
(2) To request all URIs in the triplestore which belongs to the
specified namespace.
http://localhost:8080/load:NAMESPACE
(3) To create a synonym node to link two URIs which have
the same id but different synonymous namespaces. For
example, to link omim:602080 and mim:602080 URI
together because the omim namespace is equivalent to
mim.
http://localhost:8080/sameas:NAMESPACE1-NAMESPACE2
The URLRewrite library matches the URL syntax with the appropriate
RDFizer program. This software component controls the
information workflow by interpreting rules defined by regular
expressions. Examples of rules stored in the URLRewrite configuration
file follow. The first rule below calls ncbi-pubmed2rdf.jsp, a
program that invokes the NCBI efetch utility to obtain the corresponding
PubMed document in XML format and transforms it into
RDF using XPath queries.
Original URL:
http://bio2rdf.org/jsp-bio2rdf/ncbi-pubmed2rdf.jsp?id=12728276
The rule:
<rule><from>ˆ/pubmed:(.*)</from>
<to>/jsp-bio2rdf/ncbi-pubmed2rdf.jsp?id=$1</to>
</rule>
2.5. Resulting REST-like URL
http://bio2rdf.org/pubmed:12728276
The next rule forwards to the Bio2RDF.org server any URI request
that cannot be locally resolved since there is no rdfizer program
associated to the URL namespace. This forwarding rule chains
Bio2RDF resolvers, just like DNS servers do.
<rule><from>ˆ/(.*):(.*)</from>
<to type="redirect"> http://bio2rdf.org/ 1: 2</ to>
</rule>
When a query is made to the Bio2RDF service for an unknown
URI, for example http://bio2rdf.org/biocyc:MONOMER-9282 for
which there is no rdfizer program and no URLRewrite rule, the server
responds with a graph of type bio2rdf:Unknown.
The flexible Bio2RDF approach allows the replacement of one
rdfizer by another simply by modifying the corresponding URL rewrite
rule. It is also possible to add a new rdfizer program working
with private data locally stored in a relational database. Once a
new extension is added, new knowledge sources can then be
merged with Bio2RDF. This is the way that the myBio2RDF application
learns how to explore a new knowledge space.
3. Results
The Bio2RDF project is still in development, but its RDF document
service has been made available to the scientific community.
More than twenty different public bioinformatics data
sources are now available in a normalized RDF format from the
Fig. 1. Bio2RDF knowledge system framework architecture.
710 F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716
Bio2RDF.org server. This is a knowledge space of millions of documents.
Some of the public databases were downloaded into a
MySQL database from where documents were converted by the
rdfizer programs.
These databases are NCBI’s Entrez Gene and OMIM, Kegg’s
pathway and Ligand, MGI mouse’s annotations and HGNC human’s
annotations, OBO open source ontology, PDB the Protein
Data Bank, and finally ChEBI the chemical entities database from
EBI. The MeSH RDF version of the medical vocabulary comes
from previous work done by van Assem [51]. By locally storing
these major databases, hundreds of RDF documents, related to
a specific topic, can be extracted in minutes rather than hours.
It also helps to respect NCBI usage restrictions [52]. Some other
knowledge bases are also available in RDF format from the
Bio2RDF server, although they are not hosted on it: the UniProt
protein knowledge base and its taxonomy that were recently
made available in RDF format [53], PubMed, GenBank and PubChem
from NCBI accessible with the efetch utility, Reactome
and Prosite. As an example, pathway definitions are offered in
BioPAX RDF format from the Reactome website. These documents
are accessed in real time from the usual website HTTP
service. Table 1 gives the number of RDF documents downloaded
from public databases and locally stored in our database. This
knowledge space corresponds approximately to 163 million of
well-formed RDF documents using normalized URIs and respecting
the Bio2RDF ontology.
The databases downloaded had different formats. The UniProt
knowledge base was available directly in RDF format. NCBI offered
all its main databases in ASN.1 format from its FTP site
where the Entrez Gene database was downloaded in ASN.1 format
before its conversion into XML format. Recent work by Sahoo
[54] has been done at NIH to convert Entrez Gene to RDF but this
resource is not publicly available so it cannot be used yet. The
OMIM database was available only in tabulated text files so the
efetch utility was used to extract each OMIM’s record individually
in XML format. Gene Ontology was available from three different
sources (GO, OBO and UniProt) in three different RDF schemas.
The GO’s FTP server was chosen because it was the authoritative
website. The PDB server releases all its records over an FTP server.
Kegg’s pathways were downloaded from their FTP server in KGML
(http://www.genome.jp/kegg/docs/xml/), an XML proprietary format.
The LIGAND database, with documents about compound,
reaction and enzyme, could be downloaded only in text format.
A Perl program was written to rdfize them. The MGI mouse genome’s
annotations, originally in tabulated text files, were transformed
into RDF the same way. Finally, the OBO’s ontologies
were downloaded in a RDF version.
With the Bio2RDF server or the myBio2RDF application, it is
possible to browse millions of RDF documents using the Sesame
explorer to view HTML page, Piggy Bank [32] or the experimental
FireFox extensions: LSID browser [34] or Tabulator [55]. The next
section explains how an agent can automate this process and the
role of the Elmo RDF crawler.
4. Parkinson use case
The potential of the Bio2RDF approach to build specialized
mashups is illustrated by applying it to the construction of a
knowledge base about Parkinson’s disease (PD). This disease was
chosen because it was already analyzed by the BioRDF subgroup
of the HCLS community [60], and also in reason of the availability
of a Parkinson’s disease specialist at the CHUL Research Center,
Claude Rouillard. The following paragraph explains part of his
research.
Parkinson’s disease is a slow progressive neurodegenerative disorder.
Most cases of PD are sporadic, but rare familial forms of the
disease do occur. However, the mechanisms underlying the selective
death of nigral dopamine (DA) neurons are still unknown. Nuclear
receptors constitute a conserved family of ligand activated
transcription factors regulating gene expression. We and others
have provided several lines of evidence suggesting an important
involvement of a subgroup of nuclear receptors specifically associated
with DA neurotransmission in the developing and mature
CNS. This subgroup includes the Retinoid X Receptor (RXR) and orphan
members of the thyroid/steroid nuclear receptor family
named the Nur family, which includes Nurr1, Nur77, and Nor-1.
Nurs are classified as early response genes, and are induced by diverse
signals, including growth factors, cytokines, peptide hormones,
neurotransmitters, and stress. Their ability to sense and
rapidly respond to changes in the environment seems to be a hallTable
1
Number of RDF documents from public databases available with Bio2RDF
Data source Short URI example Number of RDF documents Format of source data Hosted version
genenames.org hgnc:4922 27,634 Tabulated December 2007
informatics.jax.org mgi:96103 70,172 Tabulated June 2007, MGI 3.54 release
ncbi.nlm.nih.gov omim:146200 18,284 XML December 2007
ncbi.nlm.nih.gov geneid:3098 3,315,893 XML December 2007
genome.ad.jp path:mmu00010 68,307 KGML December 2007, Release 44.0+/12-19
genome.ad.jp cpd:C00011 15,006 Text December 2007, Release 44.0+/12-19
genome.ad.jp dr:D00001 6755 Text December 2007, Release 44.0+/12-19
genome.ad.jp ec:2.7.1.1 4,958 Text December 2007, Release 44.0+/12-19
genome.ad.jp gl:G00001 10,972 Text December 2007, Release 44.0+/12-19
genome.ad.jp rn:R00014 7422 Text December 2007, Release 44.0+/12-19
ebi.ac.uk chebi:16526 13,360 Tabulated December 2007, Release 39.0
rcsb.org pdb:1HKC 48,091 XML December 2007
geneontology.org go:0004396 24,634 OBO/RDF December 2007
nlm.nih.gov mesh:D006593 23,512 RDF Febuary 2007
obofoundry.org obo’s 54 namespaces 108,955 OBO/RDF December 2007
beta.uniprot.org uniparc:UPI00005AC213 30,261,843 RDF
beta.uniprot.org uniprot:P19367 4,177,176 RDF
beta.uniprot.org uniref:UniRef50_P19367 7,990,452 RDF
beta.uniprot.org taxon:9606 441,422 RDF
ncbi.nlm.nih.gov genbank:NP_277035 61,132,599 XML
ncbi.nlm.nih.gov pubmed:3207429 17,000,000 XML
ncbi.nlm.nih.gov pubchem:3313 38,000,000 XML
reactome.org reactome:70326 8,332 BioPAX/RDF
expasy.org prosite:PS00378 2,819 HTML
Total 162,778,598
F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716 711
mark of this subgroup. Numerous results suggest that impaired
Nurr1 function may be associated with an increased vulnerability
of dopamine neurons to degeneration in Parkinson’s disease
whereas both Nur77 and Nor-1 are important signals for apoptosis
pathways outside the brain. Interestingly, Nur77 functions as a survival
factor in the nucleus whereas it is a potent killer when migrating
to the mitochondria.
The mashup created with Bio2RDF about PD will help answer
these questions:
1. Which GO terms describe our four genes of interest (Rxr, Nurr1,
Nur77, and Nor-1)?
2. Which articles mentioning our four genes of interest are related
to apoptosis AND cytoplasm and also mention genes having GO
annotations about apoptosis OR cytoplasm?
First, a knowledge base is built to answer these questions. This
knowledge base is loaded with the relevant documents from different
sources to answer a specific question.
The data mashup building procedure can be reproduced using
the myBio2RDF application to build the needed knowledge base.
Initially, the related RDF documents are added to the triplestore
to our four genes of interest. This is done by submitting the following
URIs to the Elmo crawler application:
http://localhost:8080/bio2rdf/search:nur77@geneid
http://localhost:8080/bio2rdf/search:nurr1@geneid
http://localhost:8080/bio2rdf/search:nor-1@geneid
http://localhost:8080/bio2rdf/search:rxr@geneid
This shows the search service provided by Bio2RDF which invokes
the NCBI’s own Entrez search. The search for Nur77 returns
38 genes, Nurr1 returns 28 genes, Nor-1 returns 17 genes and RXR
returns 78 genes. With the next step, we add the PubMed and GO
annotations they are referring to are added into the triplestore by
submitting the two following URIs:
http://localhost:8080/bio2rdf/load:pubmed
http://localhost:8080/bio2rdf/load:go
With these documents in the triplestore, we can now try to answer
the questions. First, we want to characterize our genes of
interest: RXR, Nurr1, Nur77 and Nor-1. Each gene’s description
from Entrez refers to several Gene Ontology identifiers. The load:go
URI is used to fetch the identifier’s description. With a SeRQL query
we gather all these GO terms about our genes of interest and the
result is transferred to a spreadsheet to create a cross table out
of it. The query returns 385 distinct GO terms and a total amount
of 1295 annotations for all of the four genes.
Line 2 specifies the three columns of the result. The first column,
searchLabel, contains the name of the terms that were looked
for in NCBI’s Entrez. The second column contains the name of the
gene related to one of the four genes we searched for. The third column
contains the GO term name. The result of this query, based on
documents from two different sources (GO and Entrez Gene), is
shown in Table 2.
In Table 2, two GO terms, cytoplasm and apoptosis, are highlighted
because they are involved in the second question. We
choose these terms because Nur77 mediated apoptosis outside
the brain involves its translocation from the nucleus to the cytoTable
2
GO terms frequency for four genes of interest related to Parkinson’s disease
GO terms Genes of interest
Nor-1 Nur77 Nurr1 RXR Total
Nucleus [go:0005634] 7 14 12 28 61
Regulation of transcription, DNA-dependent [go:0006355] 7 10 10 21 48
Protein binding [go:0005515] 3 11 9 21 44
Metal ion binding [go:0046872] 4 9 8 21 42
Transcription [go:0006350] 5 10 7 20 42
Transcription factor activity [go:0003700] 6 7 8 21 42
Zinc ion binding [go:0008270] 4 9 6 20 39
Sequence-specific DNA binding [go:0043565] 6 6 8 18 38
Steroid hormone receptor activity [go:0003707] 4 5 5 17 31
Signal transduction [go:0007165] 3 11 6 10 30
Positive regulation of transcription from RNA polymerase II Promoter [go:0045944] 3 6 4 10 23
Cytoplasm [go:0005737] 7 4 7 18
DNA binding [go:0003677] 1 3 5 9 18
. . .
Anti-apoptosis [go:0006916] 1 2 1 4
Apoptosis [go:0006915] 3 1 4
Biological_process [go:0008150] 1 1 1 1 4
. . .
712 F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716
plasm. Once again, we can answer this more complicated question
with a SeRQL query over the same knowledge base.
Genes are the starting point of the knowledge base graph. Lines
8 and 9 show that genes are linked to GO terms and PubMed articles.
Line 6 selects genes that come from Entrez Gene. We want to
restrict the search for articles to genes having either ‘‘apoptosis” or
‘‘cytoplasm” in their GO annotations. This restriction is specified on
lines 15 and 17 where http://bio2rdf.org/go:0006915 is the URI for
‘‘apoptosis” and http://bio2rdf.org/go:0005737 is the URI for ‘‘cytoplasm”.
Finally, with line 12, we do a full text search over all text
(title, abstract and MeSH annotations) of PubMed articles for literals
specified on lines 21 and 23. The Fig. 2 illustrates the building
and the querying of the mashup. Each black node corresponds to
a building step of the knowledge base when documents were
added to the triplestore. White nodes correspond to restrictions
in the second SeRQL query.
Finally, we have transferred the query result into a spreadsheet
in order to create a last cross table. Table 3 depicts the relationship
between the genes of interest and various apoptosis-related factors.
Some of the related links are relevant to the mechanisms by
which the genes of interest might be involved in dopamine neuron
degeneration.
5. Discussion
5.1. URI normalization
In this project. we have created RDF documents from many different
sources, implementing a simple URI normalization scheme
to solve the recombinant effect described in the Material and
Method section. The good usage of URIs is a central issue in RDF
bioinformatics databases. Providers such as UniProt have now
replaced LSIDs (used in the development phase of the beta.
uniprot.org project) by HTTP URIs. Bio2RDF has adopted the same
Fig. 2. Number of documents and subsequent restrictions on the knowledge base.
F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716 713
approach and we hope to see other database providers make their
data available in RDF with a similar service based on dereferencable
URIs by HTTP queries.
5.2. Compatibility with ongoing Semantic web projects
By designing Bio2RDF according to the linked data rules [49],
we have created a knowledge space directly usable by a true
semantic web browser such as Tabulator, in order to browse the
knowledge space of bioinformatics and define the queries dynamically
based on the path traveled. Bio2RDF is in use in the demo
section of the Tabulator [56]. This demo shows the usefulness of
linked data with normalized URIs from different databases.
The Bio2RDF RDF graph can also be browsed with a LSID browser
such as [34] through a SOAP web service [57] to request the
RDF graphs by LSID. When using this service, the Bio2RDF URIs
are replaced by LSIDs with bio2rdf.org for domain name, so
http://bio2rdf.org/geneid:15275 becomes urn:lsid:bio2rdf.org:geneid:15275.
Like other LSID resolvers do, this URL returns the
corresponding graph with LSID in place of URI:
http://bio2rdf.org/urn:lsid:bio2rdf.org:geneid:15275.
Facet browsing is also an important aspect of semantic web
application interfaces. Because Bio2RDF returns an RDF graph that
can be loaded into the Piggy Bank Semantic web facet browser
[32], once a number of graphs of interest have been loaded into
its local triplestore, it is possible to do facet browsing in this
knowledge space.
Fig. 1 illustrates the use of those different browsing tools that
may be employed to browse the semantic web knowledge space
available through the Bio2RDF service.
5.3. Extendability
The Bio2RDF architecture was designed with extendability in
mind. In addition to the Bio2RDF web services, the myBio2RDF
application enables users to integrate local and private data and
link them to the Bio2RDF knowledge space. New database sources
can easily be added to the system in a few simple steps:
1. Design the RDF document representing the data, using a tool
such as Protégé;
2. Write the corresponding rdfizer program to convert the data
into a well formed RDF/XML document;
3. Install the new rdfizer program under the Bio2RDF servlet of the
myBio2RDF installation;
4. Add a rewrite rule to the urlrewrite.xml configuration file to
associate the new rdfizer program to the URI associated to the
namespace;
5. Restart the myBio2RDF servlet.
Once a new rdfizer program for a public database has been written,
it could be submitted to the Bio2RDF project team for addition
to the public Bio2RDF service.
5.4. Scalability of complexity
In the near future, more knowledge will be available to the scientific
community, from more different sources and with increasing
complexity. How will data be integrated without using a
strategy to keep complexity constant in the underlying system?
This is the most important contribution of the RDF framework,
and the most useful characteristic of a triplestore. Without a triplestore,
RDF documents are just XML records. It is inside the triplestore
that the inherent recombining characteristic of URIs
becomes available if they are normalized. The complexity of the
knowledge stored in the triplestore can grow without any extra
programming to manage it. RDF is a framework that enables a very
simple thing: scalability of the knowledge base complexity. The
Bio2RDF project proposes to keep complexity in the bioinformatics
knowledge space under control by applying this proven semantic
web approach.
5.5. Use case
With the former use case about Parkinson’s disease, we have
shown the potential of the Bio2RDF knowledge framework to build
a very specific knowledge base, the mashup. It was then queried
with SeRQL to answer some very specialized questions. The procedure
employed in the use case is versatile because, by modifying
the SeRQL query, we can search for all kinds of relations between
genes of interest and GO terms. It is also very efficient because of
the speed and simplicity by which we can gather documents from
many different linked data sources providing RDF documents.
5.6. Bio2RDF is a work in progress
The Bio2RDF’s ontology and its rdfizer programs are not definitive.
The RDF document format will still evolve. We invite interested
bioinformaticians to join the bio2rdf.sourceforge.net project.
There are many more rdfizers to be written. The bio2rdf.owl
Table 3
Article frequency of related genes, containing apoptosis and cytoplasm in their abstract, and related to our genes of interest
Genes of
interest
Related genes GO terms
Apoptosis
[go:0006915]
Cytoplasm
[go:0005737]
Total
result
Nur77 BCL2-like 11 (apoptosis facilitator) (Bcl2l11) [geneid:12125] 2 2
Cyclin-dependent kinase inhibitor 2D (p19, inhibits CDK4) (CDKN2D) [geneid:1032] 1 1
Histone deacetylase 7A (HDAC7A) [geneid:51564] 2 2
Nur77 downstream gene 1 () [geneid:368204] 1 1
Tumor necrosis factor (TNF superfamily, member 2) (TNF) [geneid:7124] 4 4
v-akt murine thymoma viral oncogene homolog 1 (AKT1) [geneid:207] 15 15 30
Nurr1 secreted phosphoprotein 1 (Spp1) [geneid:20750] 1 1
RXR B-cell leukemia/lymphoma 2 related protein A1a (Bcl2a1a) [geneid:12044] 1 1
Caspase 8, apoptosis-related cysteine peptidase (CASP8) [geneid:841] 22 22
Nuclear receptor coactivator 2 (NCOA2) [geneid:10499] 1 1
v-rel reticuloendotheliosis viral oncogene homolog A, nuclear factor of kappa light polypeptide gene
enhancer in B-cells 3, p65 (avian) (RELA) [geneid:5970]
10 10
Total result 22 53 75
714 F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716
ontology is just at an early stage of development, it now needs to
be adopted and augmented by the community, as it was the case
for the BioPAX ontology. An ontology belongs to a community
who adapts it, uses it and shares it. With the warehouse stored
into a triplestore, it is possible to query the local knowledge base
with SeRQL queries. However, the semantic web is meant to be
distributed. With more RDF resources available on the web and
by using the SPARQL [58] language and protocol, a standard defined
by the W3C, the data warehousing concept could become
obsolete in the future. This is one perspective of the semantic
web.
6. Conclusion
In the Bio2RDF project, our main goal was to create a framework
that could be used to create an on-demand knowledge base
to form a mashup of data in the bioinformatics domain. This framework
provides an access to normalized RDF documents from many
different sources, and offers a method for users to add knowledge
sources by creating new rdfizers and also a way to keep privacy of
private data by using its built-in routing capability.
We have shown that the semantic web approach for automatic
knowledge aggregation is promising. Other research projects have
explored data integration with similar approaches but Bio2RDF
showed that it is possible to scale up to millions of documents
(in our example, 163 million documents from more 20 different
data sources). With the availability of software dealing with RDF
documents, the elaboration of a friendly user interface to query
our networked data was a secondary concern, at least as a first
step. Despite the ongoing need for friendly user interfaces to the
Bio2RDF service, semantic web tools working with RDF are in rapid
evolution. Our message to the bioinformatics community is the following:
good work can already be done with current semantic web
software, and more effort should be directed to improve the quality
of RDF data.
Since we now have access to large amounts of RDF files from
biological databases, we will study the underlying graph created
by linking them together and apply Bio2RDF to knowledge discovery.
By giving access to a knowledge space with well organized
data in the semantic web of life sciences, we believe that Bio2RDF
is an example of tool that can help eliminate some of the social
hurdles (aka.creeps [59]) to the adoption of this valuable
technology.
The myBio2RDF application, which is a modified version of the
Sesame triplestore with rdfizers, can be downloaded at http://
sourceforge.net/projects/bio2rdf/.
Acknowledgments
The Bio2RDF software project was made possible because of the
availability of software from the open source community. Our first
thanks go to programmers of this community. It was possible to
create the Bio2RDF service because huge amounts of curated
knowledge are made publicly available to the biologist community
by the data providers. We also thank them, especially the curators
without whom knowledge tagging would not be a reality. We also
thank Claude Rouillard for his help in the production of the example
with Parkinson’s disease. Finally, we would like to thank the
reviewers for their valuable suggestions.
Francßois Belleau was a recipient of a studentship from Génome
Québec and Marc-Alexandre Nolin was a recipient of a studentship
from the Canadian Institutes of Health Research. This work has
been financed in part by the Atlas of Genomic Profiles of Steroid
Action, a project funded by Genome Canada and Génome Québec.
This paper is an extension of our workshop paper’Bio2RDF: Towards
a Mashup to Build Bioinformatics Knowledge System’
published in WWW2007/HCLS-DI (http://www2007.org/work-
shop-W2.php).
References
[1] Schuler GD, Epstein JA, Ohkawa H, Kans JA. Entrez: molecular biology database
and retrieval system. Methods Enzymol 1996;266:141–62.
[2] Fujibuchi W, Goto S, Migimatsu H, Uchiyama I, Ogiwara A, Akiyama Y, et al.
DBGET/LinkDB: an integrated database retrieval system. Pac Symp Biocomput
1998:683–94.
[3] Fox JA, McMillan S, Ouellette BFF. A compilation of molecular biology web
servers: 2006 update on the Bioinformatics Links Directory. Nucleic Acids Res
2006;34:W3–5.
[4] http://www.pathguide.org/.
[5] Stein LD. Integrating biological databases. Nat Rev Genet 2003;4:337–45.
[6] http://www.w3.org/RDF/.
[7] http://www.w3.org/2004/OWL/.
[8] Aduna Sesame, http://www.openrdf.org.
[9] http://en.wikipedia.org/wiki/Mashup_(web_application_hybrid).
[10] Davidson SB, Overton C, Buneman P. Challenges in integrating biological data
sources. J Comput Biol 1995;2:557–72.
[11] Köhler J, Philippi S, Lange M. SEMEDA: ontology based semantic integration of
biological databases. Bioinformatics 2003;19:2420–7.
[12] Kanehisa M, Goto S, Hattori M, Aoki-Kinoshita KF, Itoh M, Kawashima S, et al.
From genomics to chemical genomics: new developments in KEGG. Nucleic
Acids Res 2006;34:D354–7.
[13] Joshi-Tope G, Gillespie M, Vastrik I, D’Eustachio P, Schmidt E, de Bono B, et al. .
Reactome: a knowledgebase of biological pathways. Nucleic Acids Res
2005;33:D428–32.
[14] Life Sciences Identifier, http://www.omg.org/cgi-bin/doc?lifesci/2003-
12-02.
[15] Stevens R, Baker P, Bechhofer S, Ng G, Jacoby A, Paton NW, et al. TAMBIS:
transparent access to multiple bioinformatics information sources.
Bioinformatics 2000;16:184–5.
[16] BioPAX: Biological Pathway Exchange, http://www.biopax.org.
[17] The UniProt Consortium. The Universal Protein Resource (UniProt). Nucl. Acids
Res 2007; 35: D193-D197.
[18] Cheung KH, Yip KY, Smith A, Deknikker R, Masiar A, Gerstein M. YeastHub: a
semantic Web use case for integrating data in the life sciences domain.
Bioinformatics 2005;21(1):i85–96.
[19] Shaban-Nejad A, Baker C, Haarslev V, Butler G. The FungalWeb ontology:
semantic web challenges in bioinformatics and genomics. The Semantic Web—
ISWC 2005, 2005;3729:1063–1066.
[20] http://dev.isb-sib.ch/projects/uniprot-rdf/migration.html.
[21] ftp://ftp.uniprot.org/pub/databases/uniprot/current_release/rdf/core.owl.
[22] The Protégé Ontology Editor and Knowledge Acquisition System, http://
protege.stanford.edu/.
[23] Baker C, Su X, Butler G, Haarslev V. Ontoligent Interactive Query Tool.
Semantic Web Beyond Comput Hum Exper 2006;2:155–69.
[24] http://www.racer-systems.com.
[25] Stephens S, LaVigna D, DiLascio M, Luciano J. Aggregation of
bioinformatics data using semantic web technology. J Web Semantic
2006:4.
[26] Hamosh A, Scott AF, Amberger JS, Bocchini CA, McKusick VA. Online Mendelian
inheritance in man (OMIM), a knowledgebase of human genes and genetic
disorders. Nucleic Acids Res 2005;33:D514–7.
[27] Maglott D, Ostell J, Pruitt KD, Tatusova T. Entrez gene: gene-centered
information at NCBI. Nucleic Acids Res 2007;35:D26–31.
[28] Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, et al. Gene
ontology: tool for the unification of biology. The Gene Ontol Consortium Nat
Genet 2000;25:25–9.
[29] Louie B, Mork P, Martin-Sanchez F, Halevy A, Tarczy-Hornoch P. Data
integration and genomic medicine, J Biomed Inform, 40(1), Bio*Medical
Informatics, February 2007, p. 5–16.
[30] http://java.sun.com/products/jsp/jstl/.
[31] http://tuckey.org/urlrewrite/.
[32] Huynh D, Mazzocchi S, Karger D. Piggy bank: experience the semantic web
inside your web browser, International Semantic Web Conference (ISWC)
2005.
[33] http://simile.mit.edu/welkin/.
[34] http://lsids.sourceforge.net/resources/firefox-lsid-browser/.
[35] Gruber T. Toward principles for the design of ontologies used for knowledge
sharing. Int J Hum Comput Stud 1995;43:907–28.
[36] http://bio2rdf.org/bio2rdf-2007-02.owl.
[37] The Dublin Core Metadata Initiative, http://dublincore.org/.
[38] Brickley D, Miller L. FOAF Vocabulary Specification, http://xmlns.com/foaf/
spec/.
[39] Knouf N. bibTeX Definition in Web Ontology Language (OWL) Version 0.1,
Working Draft. http://zeitkunst.org/bibtex/0.1/, 2004.
[40] http://simile.mit.edu/RDFizers/.
[41] http://eutils.ncbi.nlm.nih.gov/entrez/query/static/efetch_help.html.
[42] Hubbard TJ, Aken BL, Beal K, Ballester B, Caccamo M, Chen Y, et al. Ensembl
2007. Nucleic Acids Res 2007;35:D610–7.
F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716 715
[43] Hulo N, Bairoch A, Bulliard V, Cerutti L, De Castro E, LangendijkGenevaux
PS, et al. The PROSITE database. Nucleic Acids Res
2006;34:D227–30.
[44] Thomas Fielding R. Architectural Styles and Design of Network-based Software
Architectures, PhD Thesis, University of California, 2000.
[45] http://www.w3.org/DesignIssues/LinkedData.
[46] Eppig JT, Bult CJ, Kadin JA, Richardson JE, Blake JA, Anagnostopoulos A, et al.
Mouse genome database group. The mouse genome database (MGD): from
genes to mice—a community resource for mouse biology. Nucleic Acids Res
2005;33:D471–5.
[47] HUGO Gene Nomenclature Committee (HGNC), Department of Biology,
University College London, Wolfson House, 4 Stephenson Way, London NW1
2HE, UK, http://www.genenames.org/.
[48] http://www.berkeleybop.org/ontologies/.
[49] Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, Weissig H,
et al. The protein data bank. Nucleic Acids Res 2000;28:
235–42.
[50] Degtyarenko K, Matos PD, Ennis M, Hastings J, Zbinden M, McNaught A, et al.
ChEBI—Chemical Entities of Biological Interest. Nucleic Acids Res, Database
Summary Paper 646.
[51] van Assem M, Malaisé V, Miles A, Schreiber G. A method to convert thesauri to
SKOS. Semantic Web Res Appl 2006:95–109.
[52] NCBI User system requirements, http://eutils.ncbi.nlm.nih.gov/entrez/query/
static/eutils_help. html#UserSystemRequirements.
[53] http://beta.uniprot.org.
[54] Sahoo SS, Bodenreider O, Zeng K, Sheth A. An experiment in integrating large
biomedical knowledge resources with RDF: Application to associating genotype
and phenotype information, http://www2007.org/workshops/paper_149.pdf.
[55] Berners-Lee T, Chen Y, Chilton L, Connolly D, Dhanaraj R, Hollenbach J, et al.
Tabulator: exploring and analyzing linked data on the semantic web. In:
Proceedings of the 3rd International Semantic Web User Interaction Workshop
(SWUI06) workshop, Athens, Georgia, 6 November 2006.
[56] http://dig.csail.mit.edu/2007/tab/tabtutorial.html.
[57] http://bio2rdf.org/authority.
[58] http://www.w3.org/TR/rdf-sparql-query/.
[59] Good BM, Wilkinson MD. The life sciences semantic web is full of creeps! Brief
Bioinform 2006;7:275–86.
[60] BioRDF subgroup of the HCLS community, http://www.w3.org/2001/sw/hcls/.
716 F. Belleau et al. / Journal of Biomedical Informatics 41 (2008) 706–716