Transcriptional regulation of gene expression is an important mechanism in many biological processes. Aberrations in this mechanism have been implicated in cancer and other diseases. Effective investigation of gene expression mechanisms requires a system-wide integration and assessment of all available knowledge of the underlying molecular networks. This calls for a method that effectively manages and integrates the available data. We have built a semantic web based knowledge system that constitutes a significant step in this direction: the Gene Expression Knowledge Base (GeXKB). The GeXKB encompasses three application ontologies: the Gene Expression Ontology (GeXO), the Regulation of Gene Expression Ontology (ReXO), and the Regulation of Transcription Ontology (ReTO). These three ontologies, respectively, integrate gene expression information that is increasingly more specific, yet decreasing in coverage, from a variety of sources. The system is capable of answering complex biological questions with respect to gene expression and in this way facilitates the formulation or assessment of new hypothesis. Here we discuss the architecture of these ontologies and the data integration process and provide examples demonstrating the utility thereof. The knowledge base is freely available for download and can be queried through a SPARQL endpoint (http://www.semanticsystems-biology.org/apo/).
Research in the Life Sciences is supported by a plethora of
databases (see overview at www.pathguide.org). Moreover,
the continuing advancements in functional genomics
technologies make it possible to create an overwhelming
amount of data in a single experiment. The many
hypotheses that can be derived from such experiments must
be assessed against a multitude of information and
knowledge bases, often represented in a variety of formats.
Scientists therefore become increasingly dependent on
sophisticated computer technologies to integrate and
manage all the available information. Furthermore, the
drastic increase in the available information and a lack of
adhering to accepted formal representations across all
disparate knowledge bases allows only a fraction of the
knowledge to be easily considered in the analysis of new
data, or causes a user to query many databases individually,
sometimes even without the support of ontology terms that
would warrant a common semantics of queries in different
databases. As discussed by
Need for an integrated resource that captures gene expression knowledge
on a large variety of cellular processes that control the timing and level of transcription of an individual gene, often in a cell- or condition specific manner. Regulation of the expression of protein coding genes is extensively studied.
tion and translation. During the process of transcription, proteins called transcription factors bind to specific DNA sequence motifs (binding sites) of a gene, playing a key role in initiating or inhibiting the formation of an active RNA Polymerase II transcription complex. Active transcription produces pre-mRNAs which are subsequently processed (removal of introns, and polyadenylation of the transcript) upon which mature mRNAs are transported from the nucleus to the cytoplasm where the mRNA is translated into a protein. Regulatory processes of gene expression occur at different levels, enabling the cell to adapt to different conditions by controlling its structure and function. Furthermore, the process of gene expression may also be influenced at the epigenetic level, where nucleotide or protein modifications can cause heritable changes in expression of otherwise identical gene sequences. Abnormalities in the regulation of gene expression can cause diseases such as the occurrence of malignant cell proliferation.
for a systems-wide understanding of gene regulation, there
is a need for efficiently capturing knowledge of this domain
in its entirety and to further facilitate efficient querying of
this data. For instance, the complex one-to-many
relationships of a transcription factor like Myc includes thousands
of target genes, representing a wide variety of functions and
processes. An ontology-driven approach would best solve
the issue of knowledge querying, representation and
management. Previously, attempts have been made to model the
gene regulation process; resulting in the Gene Regulation
Ontology (GRO)
relevant for the domain of gene expression to support the discovery of new scientific knowledge. We have named this knowledge system: the Gene Expression Knowledge Base (GeXKB). This system is conceived as part of the Semantic
biology.org) initiative and comprises at the current stage three application ontologies that capture the knowledge about gene expression, namely the Gene Expression
2
GeXKB is designed to provide the molecular biologist with a knowledge system that captures knowledge on a variety of aspects of the gene expression process. To this end it should be able to provide answers to questions like: • ‘Which are the proteins that act as chromatin remodeling proteins and as modulators of transcription factor activity?’ • ‘Which are the proteins that participate in two successive regulatory pathways?’. • ‘Which are the transcription factors (Human) that are located in the cytoplasm?’.
The following design principles were followed in the process of GeXKB development: • 'is a' completeness • 'all-some' semantics • only classes used for modelling of the domain of discourse (see Table 1) • maximal flexibility both for users and for future extensions 3
number of well established biomedical ontologies, first of
all GO
3.1
Data integration pipeline
AND
integration. It was developed on the basis of the Science
ticscience/wiki/SIO) with the addition of few terms from other ontologies. The origin of the terms is preserved in external references. The ULO is generated on the fly by the pipeline and does not exist as an individual artifact. The upper level term IDs are of the form ‘SSB:nnnnnnn’.
fragments of Biological Process, complete Cellular
type' branch), and the Biorel ontology (Blondé et al. 2011).
GeXO, the term ‘gene expression’ (GO:0010467) and all its descendants are imported. For ReXO and ReTO the corresponding GO terms are: 'regulation of gene expression' (GO:0010468) and 'regulation of transcription, DNA dependent' (GO:0006355). We refer to these three terms as sub-roots. Each of them is connected to the ULO as a subclass of 'biological process'. To ensure 'is a' completeness, each of the ontologies is complemented with an auxiliary term - (‘gene expression process’ (GeXO:0000001), 'process of regulation of gene expression’ (ReXO:0000001), ‘process of regulation of DNA-dependent transcription’ (ReTO:0000001)), which becomes the parent of all the terms that did not have an 'is a' path to the subroot. Apart from this, the three seed ontologies are structurally identical (Figure 1). 3.1.2 Building species specific intermediate ontologies:
Homo sapiens, Mus musculus and Rattus norvegicus. The corresponding three species-specific intermediate ontologies were developed in the following steps: (1) For each species GOA annotations are used to extract all the associations involving domain specific Biological Process terms incorporated in the previous phase. The corresponding proteins are added as child terms to the upper level term ‘protein’ (SSB:0001211) and referred to as 'core proteins' hereafter. (2) From the IntAct database all the interactions involving at least one of the core proteins are retrieved and incorporated into the knowledge base along with their pertinent information. This results in a further extension of the set of proteins in the KB.
This is the final phase in the generation of the ontologies which proceed as follows: (1) The species specific ontologies (from the previous step) are merged together. (2) From the KEGG database all the pathways involving at least one of the core proteins are extracted and incorporated in the KB along with the pertinent information. The pathway terms become children of the term 'SSB:0011221' ( 'pathway', 'BioPAX:Pathway'). The corresponding KEGG orthology groups are incorporated as children of the term 'protein cluster' (SSB:0001122). This step results in a second extension of the set of proteins. (3) Putative orthology relationships were computed with the use of the high-performance library
C++ implementation of the OrthoMCL algorithm
3.1.4 Enhancing the utility of the ontologies: (1) Transitive closures were constructed with the use of the library ONTO-PERL for the following relation types: 'is a', 'part of', ‘regulates’. (2) The ontologies were exported in a number of formats:
RDF, OWL, XML, and DOT. (3) The RDF exports were used to populate a triple store, refer Table 2 (Virtuoso Open Link).
Ontology GeXO
extension of the WWW which aims at building a web of
data accessible both by computers and human beings. This
new technology is increasingly gaining momentum, in
particular in the domain of Life Sciences
versions of the ontologies have been loaded into Open Link
Virtuoso (http://virtuoso.openlinksw.com) and can be
accessed via a SPARQL query page
(http://www.semanticsystems-biology.org/apo/queryingcco/sparql). In contrast to
other Semantic Web formalisms, such as OWL, RDF
enables handling of large amounts of knowledge due to its
simple and flexible syntax, making querying tractable.
However, on the downside the low expressivity of RDF/RDFS
imposes limitations on the inferencing over the knowledge
base. To overcome this limitation,
tic Web is the quick expansion of the Linked Data cloud
migration of GeXKB eventually to the Linked Data cloud.
sible. If the original IDs include a name-space (e.g. GO, MI) they were adopted without any modifications, otherwise the IDs were prepended with a name-space (for example UPKB for UniProtKB or NCBIgn for NCBI Gene), separated by a colon from the original ID (the colons are replaced with underscores in the RDF renderings). The re-use of the IDs benefits as well the users due to faster query execution and the familiarity of the IDs. Furthermore, in compliance with the Linked Data recommendations we minted the URIs in our own common name-space: http://www.semanticsystems-biology.org/ and have consistently used rdfs:label properties to aid human readability of the results.
RDF graphs No. of triples
GeXO ~3.3 million
GeXOtc ~23 million
ReXO ~3 million
ReXOtc ~19.9 million
ReTO ~2.8 million
ReTOtc ~19.1 million In this section we demonstrate the utility of GeXKB with the help of a few example SPARQL queries. These queries are available as a part of a list of sample queries provided on the query page (http://www.semantic-systemsbiology.org/apo/queryingcco/sparql). To query GeXKB, the base URI and the prefixes are set and the SELECT block specifies the variables to be part of the solution. The RDF triple pattern queried is defined in the WHERE block. The queries are as follows:
Biological question: Which proteins can act as chromatin remodeling proteins and as modulators of transcription factor activity? SPARQL query: BASE <http://www.semantic-systems-biology.org/> PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#> PREFIX ssb: <SSB#> PREFIX taxon: <SSB#NCBItx_9606> PREFIX graph1: <ReXO> PREFIX graph2: <ReTO-tc> SELECT distinct ?protein_id ?protein_name WHERE { GRAPH graph1: { ? protein_id ssb:is_a ssb:SSB_0001211 . ?b_process ssb:is_a ssb:GO_0040029 . ?b_process ssb:has_participant ? protein_id . ? protein_id ssb:has_source taxon: . } GRAPH graph2: { ssb:GO_0034401 ssb:has_participant ? protein_id . ? protein_id rdfs:label ?protein_name . } } LIMIT 4 Q2: Biological question: Which proteins participate in both the JAK/STAT signaling pathway and Apoptosis? SPARQL query: BASE <http://www.semantic-systems-biology.org/> PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#> PREFIX ssb: <SSB#> PREFIX taxon: <SSB#NCBItx_9606> PREFIX pathway1: <SSB#KEGG_ko04630> PREFIX pathway2: <SSB#KEGG_ko04210> PREFIX graph: <GeXO> SELECT distinct ?protein WHERE { GRAPH graph: { ?prot_id ssb:is_a ssb:SSB_0001211 . ?prot_id ssb:is_member_of ?cluster . pathway1: ssb:has_agent ?cluster . ?prot_id ssb:has_source taxon: . } GRAPH graph: { ?prot_id ssb:is_member_of ?cluster . pathway2: ssb:has_agent ?cluster . ?prot_id rdfs:label ?protein . } } Q3: Biological question: Which are the transcription factors (Human) that are located in the cytoplasm? SPARQL query: BASE <http://www.semantic-systems-biology.org/> PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#> PREFIX ssb: <SSB#> PREFIX taxon: <SSB#NCBItx_9606> PREFIX location: <SSB#GO_0005737> PREFIX graph: <ReTO-tc> SELECT distinct ?protein ?protein_name WHERE { GRAPH graph: { ?protein ssb:is_a ssb:SSB_0001211 . ?protein rdfs:label ?protein_name . ssb:GO_0006355 ssb:has_participant ?protein . ?protein ssb:has_function ?function . ?function ssb:is_a ssb:GO_0003700 . location: ssb:contains ?protein . ?protein ssb:has_source taxon: . } } These queries offer just a glimpse of the repertoire of biological question that can be addressed to the knowledge system. In addition, users could also query the knowledge base in combination with other complementary semantic web resources to formulate advanced queries for hypothesis generation. This could be performed through the query federation features that are included in the latest version of
Protein ID http://www.semantic-systemsbiology.org/SSB#UPKB_Q9NS37 http://www.semantic-systemsbiology.org/SSB#UPKB_P14373 http://www.semantic-systemsbiology.org/SSB#UPKB_Q62158 http://www.semantic-systemsbiology.org/SSB#UPKB_P17947 Protein Name ZHANG_HUMAN TRI27_HUMAN TRI27_MOUSE SPI1_HUMAN
The drastic increase in the amount of data generated in the field of molecular biology and biomedicine requires efficient knowledge management practices. Ontologies certainly provide a robust method to integrate data and efficiently represent specific (sub) domain knowledge. With the creation of GeXKB, we have built a knowledge system that specifically supports researchers focusing on various aspects of gene expression. The three ontologies provide the user with the flexibility of choosing an ontology depending on the breadth and specificity of information needed. Further flexibility is afforded by a range of available formats for knowledge representation (OBO, RDF, OWL), data exchange (XML), and visualisation (DOT).
The presented examples demonstrate the utility of our knowledge base with respect to answering realistic domain specific questions, and this utility is expected to grow with its further development. The primary goal will be to augment the knowledge base with additional high quality, curated sources of information with documented transcription factor function and relations between transcription factors and their target genes.
ence and Technology (NTNU), Norway. AV was funded by