In recent years, a large amount of “-omics” data has been produced. However, these data are stored in many different species-specific databases that are managed by different institutes and laboratories. Biologists often need to find and assemble data from disparate sources to perform certain analyses. Searching for these data and assembling it is a time-consuming task. The Semantic Web helps to facilitate interoperability across databases. A common approach involves the development of wrapper systems that map a relational database schema onto existing domain ontologies. However, few attempts have been made to automate the creation of such wrappers. We developed a framework, named BioSemantic, for the creation of Semantic Web Services applicable to relational biological databases. This framework makes use of both Semantic Web and Web Services technologies and can be divided into two main parts: (i) the generation and semi-automatic annotation of an RDF view; and (ii) the automatic generation of Semantic Web Services. We have used our framework to integrate genomic data from different plant databases. BioSemantic is a framework designed to speed the development of Semantic Web Services for existing relational biological databases. Currently, it creates and annotates RDF views that enable the automatic generation of SPARQL queries. Web Services are also created and deployed automatically, and the semantic annotations of our Web Services are added automatically using SAWSDL attributes. BioSemantic is downloadable at http://southgreen.cirad.fr/?q=content/Biosemantic.
We will detail below the entire process for generating a BioSemantic SWS, which can be divided into two main parts: (i) the generation and semi-automatic annotation of an RDF view (Fig. 1); and (ii) the automatic generation of the Semantic Web Service (Fig. 2). 2
A local RDF view of the database schema is automatically created for each relational database to be integrated. Then the RDF view has to be manually annotated by experts with terms from existing bio-ontologies. The RDF views, both created and annotated, are stored in a RDF repository (Fig 1). 2.1
The research in the domain of mapping between databases and ontologies is very active and corresponds to
various motivations and approaches [
The RDF view generated by D2RQ contains the elements of the database schema: entities, attributes, keys (primary, foreign) and metadata, such as the database driver and host. The data contained in the relational databases are not included in the RDF view. Consequently, both the D2RQ API and the RDF view are requested when the data are accessed through SPARQL queries.
In the RDF view, the database schema is represented by a graph. Each node corresponds to an entity or attribute in the database, and each edge defines a relationship between two nodes. In RDF format, namespaces are used to uniquely identify each node. Namespaces provide a prefix for each node name. For example, the map:marker node indicates the “marker” concept from the “map” vocabulary. 2.3
The BioSemantic API automatically detects specific information related to the relational database schema and translates it into new properties that can be integrated into the RDF view. These metadata are then used for SPARQL query generation. This step can be seen as a semantic enrichment of the RDF view.
For this purpose, we have developed an algorithm that detects association tables:
pk= primary key of R fk= foreign keys of R } }
R is an association table { {
We can also detect the arity of association tables, i.e., the number of foreign keys they possess. The algorithm labels association tables in the RDF view with the dr:associatedTo property and indicates the arity with the dr:arity property .
There are many ways to transform inheritance relationships from an object-oriented conceptual model to a
relational model [
The annotation of the RDF view is performed manually using a text editor and must be conducted by an expert familiar with the database and/or bio-ontology.
Semantic annotations are used to select inputs and outputs of a query. We are able to find a path in one RDF view by linking the inputs to the outputs. If such a path is found in the RDF view, it is used to create a SPARQL query. To automate the creation of SPARQL queries, we implement an algorithm that is a single-pair variant of the shortest-path algorithm. Given an input graph, a source node and a destination node, it returns a path linking the two nodes through the graph. We add conditions to our shortest-path algorithm according to the types of relationships between the nodes, which can be either of the following: (i) relationships corresponding to association tables; or (ii) relationships resulting from inheritance, aggregation or composition in an objectoriented conceptual model. These conditions correspond to the metadata added to the RDF view during the automatic semantic enrichment step by the BioSemantic API.
The Web Services developer selects the bio-ontological terms to be used as input/output (Fig. 2). All of the mapping files, which are stored in the mapping file repository, are automatically parsed to find a path linking the input and output ontological terms. If such a path is found, it is used to create a SPARQL query. The query is integrated into a semantic Web Service that is then registered in a Web Service registry, such as BioCatalogue.