Given the heterogeneous nature of biological data and their intensive use in many tools, in this paper we propose a semantic data access and integration (DAI) service, based on the Grid paradigm, for the bioinformatics domain. This service uses ontologies for correlating different data sets. The DAI proposed in this work is a fundamental component of the ProGenGrid system, a grid-enabled platform, which aims at the design and implementation of a virtual laboratory where e-scientists could simulate complex “in silico” experiments, composing some popular analysis and visualization tools (e.g. Blast and Rasmol) available as Web Services, into a workflow. The main goal of the DAI is to provide bioinformatics tools with advanced functionalities and data integration services for heterogeneous biological data banks, such as PDB and Swiss-Prot. A case study of our specialized data access service for locating similar protein sequences is presented.
Keywords: Bioinformatics, DAI, Ontologies, Web Services, Computational Grid, Grid Portal, Globus Toolkit.
Complete genome sequences and protein-coding gene sets are becoming available for a growing number of organisms. While these are proving highly informative and invaluable for studying those and related organisms, at the same time they make it clear how far we still have to go before reaching an in-depth understanding of how a genome determines the lifestyle of an organism.
The increasing amount and complexity of biological
data makes it increasingly difficult to access and analyse
the data. These data, stored in different geographically
spread repositories, are heterogeneous when we consider
genomic, cellular, structure, phenotype and other types of
biologically relevant information [
The semantic relation among these data repositories is a key factor for integration in bioinformatics since it could allow a unique front end for accessing them, as required by many biological applications. Ontology could help here to localise the right type of concept to be searched for as opposed to identification of a mere label naming a search table. It includes definitions of basic concepts in the domain and relations among them, which should be interpretable both by machines and humans.
Moreover, biological repositories are often quite large
and need to be updated for annotations or when we add
new entries. To date, many tools exist for simulating
complex “in silico” experiments, that is simulations
carried out using biological data, as opposed to “in vitro”
or “in vivo” ones that are conducted respectively outside
or inside a living organism or cell. These tools need to
access heterogeneous data banks, distributed on a wide
area, and in particular need a supporting infrastructure for
obtaining successfully a result [
SRS is the most widely used data integration system for biological, biochemical and biomedical databases. It enables users of all backgrounds to intuitively access data and permits internal data to be merged with data from the public domain. The most prominent public server at EBI (http://srs.ebi.ac.uk) currently holds more than 130 biological databases. A key problem with the current structure of SRS is that it is designed only for accessing local databases. This requires the SRS administrators to provide local copies of all the databases and keep these local copies continuously up to date.
This approach uses interconnected heterogeneous databases via web hypertext links at the level of individual data items. Data retrieval in such system takes place by using the results of one query to link and jump to a particular entry in the same or another data source. However most of the potential links among data in digital form are not readily available because the relevant data, when they exist, are in different databases. In addition, each database is typically based on different and incompatible database technologies and uses different languages and vocabularies to access data. These incompatibilities are especially significant when nontextual data, such as 3D images of protein structures, accessed by author-specified keywords, need to be linked with nucleotide sequences in other databases. Because each database is typically created as a standalone application to support one functionality, linking among databases is most often an afterthought. It is possible (using an integrated approach which considers the semantic meaning of data) to dynamically create links such as a search engine.
To date, a (de facto) specialized data access service for bioinformatics, able to provide access to data and distributed tools, does not exist (yet).
A data access service is involved in many biological
experiments where Workflow techniques are needed to
assist the scientists in the design, execution and
monitoring of them. Workflow Management Systems
(WFMSs) support the enactment of processes by
coordinating the temporal and logical order of the
elementary process activities and supplying the data,
resources and application systems necessary for the
execution [
The Grid [
So, bioinformatics platforms need to offer powerful and high level modelling techniques to ease the work of escientists, as for instance exploiting Computational Grids transparently and efficiently.
ProGenGrid (Proteomics and Genomics Grid) [
In this paper we focus on the functions and architecture of a Data Access and Integration (DAI) service and its use inside the ProGenGrid platform. The use of the proposed DAI service in an experiment of searching similarity matching among proteins is presented. The outline of this paper is as follows: in Section 2, we describe the features of a bioinformatics DAI. In Section 3 we describe our DAI solution whilst in Section 4 we show the role of the DAI in the ProGenGrid system. We conclude the paper in Section 5. 2.
The interconnection of computers using Grid middleware enables the user to utilize computing power and retrieve information from heterogeneous and distributed sources transparently and efficiently. A Computational Grid could be a solution to many bioinformatics issues because it allows the deployment, distribution and management of needed biological software components, the harmonized standard integration of various software layers and services, a powerful, flexible policy definition, and control and negotiation mechanisms for a collaborative grid environment. This could reveal useful information for understanding the complex interrelation between genetic information and hereditary diseases and hence can lead to important discoveries in life science.
Bioinformatics Grids are environments built for the
specific domain of biology including hardware and
software resources needed for solving issues related to
biological experiments and simulations. Some examples
of Bioinformatics Grids are Asia Pacific BioGRID [
Web services describe an emerging XML-based distributed computing paradigm that differs from other approaches such as CORBA and Java RMI. The basic idea is to build a system out of existing Internet-based standards. Web services define the description of how to invoke service components, a protocol for conveying remote procedure calls (RPC, but also Document style Web services can be used), and the discovery mechanism for locating the service definition of relevant service providers. Web Services technology allows independence from platforms/programming languages and reusability of the code.
Data access and integration service include key steps in the data life cycle process, such as data creation and acquisition, use, modification, archiving and disposal. This process involves many data banks (data providers) and users/applications, which use the data. Coupling the Grid framework and Web Services makes it possible to build a bioinformatics DAI service satisfying the following features:
Accessibility: ease of use, support for multiple data models and database abstractions; using a Grid framework it is possible to access a large set of resources and data efficiently. Through easy to use user interfaces that hide the complexity of accessing the Grid (the so called Grid Portals), the user can access a variety of grid services.
data storage capacity, for the archival process, including space for expansion and annotation of the database; a Grid offers huge amount of data storage capacity and efficient mechanisms to move the data between grid nodes.
Intellectual property, privacy and security: the first regards ownership of sequence data, images, and other data stored in and communicated through the database, the second is the provision for preserving confidentiality of data and the last is the limit on user access. Each user is recognized in a grid infrastructure through proper credentials to access her own data or run applications on the grid. Through a single sign-on the user at first authenticates herself and then uses the resources for which she has permission rights (authorization process).
Interfaces: connectivity with other databases and applications; these represent the Web service interface to databases and application tools and are used either by the user or another service to send a query, to insert the parameters needed for the execution of a specific application and to obtain the results.
services technology it is possible to build platform independent components;
Performance: access time and data throughput; in
particular using the GridFTP [
However, there are other important issues of bioinformatics DAI that Grid and Web Service do not support such as:
Metadata Management: it includes the design, implementation, and maintenance of the metadata associated to different data sets whose semantic meaning is described through a data dictionary or ontology;
Multiple data formats: support for various data formats such as flat file, FastA and XML;
Data input support: hardware, software, and processes involved in feeding data into the database, from keyboard and voice recognition to direct instrument feed and the Internet;
and exporting data to and from different file formats;
Indexing: indexing methodology, including selection and use of the most appropriate controlled vocabulary;
Query Language: proprietary or standard query language for supporting complex query.
In the next Section, we will discuss our solution for an efficient DAI.
Our DAI has been studied for supporting integration of biological data sources and high throughput applications such as Blast or Drug design applications. It is also responsible for mapping high level requests (user requests) to low level queries, specific for each data source. These ones are in general not structured. In the following part we describe in detail this service.
The main goal of data integration is to develop the
technology to grant a user access to multiple information
systems, to retrieve information and to perform
computations transparently as if they were a single
source. The first complexity in achieving this goal is that
the information sources are often independent and
autonomous, they have completely different scheme
structures and use different data formats. To provide
uniform access, an integration system must therefore face
the problem of data heterogeneity at the system, syntax
and structural level. Moreover there is a significant
degree of semantic heterogeneity among different
information sources. Unfortunately, the semantics of
different data sources is hidden or unclear. The
integration system [
Another approach involves a data warehouse which combines data from a variety of databases in one physical location. It is very powerful for running queries against high volumes of data but it requires complex procedures for designing a global scheme and updating data.
The model that we propose is an extension of the
middleware mediator approach [
There are mediator systems that provide a semantic
bridge across information sources in complex application
domain such as biology such as TAMBIS [
In this paper, we present an information integration system that follows the mediator architecture but extends it by incorporating domain specific bioinformatics knowledge in a grid environment.
As can be seen in Figure 2, our system is made of: • Semantic Wrapper (SW), built on top of a data source, it includes i. Scheme, i.e. the (ER – Entity/Relation - or
UML) data model of a source; ii. Ontology, that describes a specific data source; iii. Relations/associations, between the local ontology and the scheme; iv. APIs, for retrieving a specific attribute or field. • Mapper, a catalog that gathers the schemes and their description coming from each SW; it is used to identify the data source of a query and to select the appropriate wrapper; • Data Source Ontology (DSO): it virtualises data sources and maps the semantic links between them; • Mediator which i) given a user query, searches semantic relations in the DSO and ii) consults the Mapper, reformulating the query, and splitting it into sub-queries, each one specific to a data source.
Regarding the Scheme (point i.), we have analysed the Swiss-Prot database (Figure 3 shows an entry) and we have built its E/R model. In particular some entities (Figure 4) involved in the scheme are: • Entry: composed of ID (corresponds to ID – IDentification - tag of Swiss-Prot), length (sequence length which is the last field of ID tag, 262 in the example of Fig. 4), seq (SQ involves the sequence i.e. TTCCP …), Descr (DE tag description), AC (AC tag - accession number), CodGen (GN tag – codifying gene), Keyw (KW tag – keywords) fields; • Taxonomy: involves ID, Name (OC tag - organism taxonomy), Synonymous (OX tag - taxonomy through cross reference) fields; • Reference: comprises ID, Title, Year, Volume and Journal (RN,RP,RC,RX,RA,RT,RL tags contain the bibliographic reference) fields.
With regard to the ontology related to each data source (point ii.), it contains semantic relations between concepts described in the data source. In particular Figure 5 shows a fragment of the ontology for Swiss-Prot, where some features for each protein (e.g. taxonomy, function etc.) are mapped. It is worth noting here that in this database some information are correlated, so using E/R scheme and the ontology it is possible to try all of the relations among data.
A possible relation among data obtained by scheme and ontology ties together entry and taxonomy with associated IDentry and IDTaxonomy (point iii.). So IDTaxonomy corresponds to the organism terms in the ontology.
We would like to integrate the following databases:
• Structure: PDB and CATH [
To build the SW component, we need to model each
data source using a ER model and an ontology. In
particular, we plan to use Gene Ontology [
Indeed, for each analysed wrapper we have
implemented in C language some functions that translate
the data source into an XML scheme and carry out the
ingestion of the entire database in our relational data
model. These features have been provided jointly with the
GRelC library [
ID GRAA_HUMAN STANDARD; PRT; 262 AA. AC P12544; DT 01-OCT-1989 (Rel. 12, Created) DT 01-OCT-1989 (Rel. 12, Last sequence update) DT 01-OCT-2004 (Rel. 45, Last annotation update) DE Granzyme A precursor (EC 3.4.21.78) GN Name=GZMA; Synonyms=CTLA3, HFSP; OS Homo sapiens (Human).
OC Eukaryota; Metazoa; Chordata; Craniata; Vertebrata.
OX NCBI_TaxID=9606;
RN [
CC -!- FUNCTION: This enzyme is necessary for target cell CC lysis in cell- mediated immune responses. It cleaves after
The Mapper contains a catalogue of data source schemes and a brief description. It is worth noting here that it contains the logical file name of the scheme associated with one or more physical file names (for instance EMBL databank has a relational, flat file and XML version corresponding each to a Mapper entry).
Data Source Ontology (DSO) classifies the data sources w.r.t. some features providing a unified conceptual level representation of its registered component resources.
In the following text we show how concepts in different ontologies are linked. As an example, the relation “polypeptide_chain(is_composed, SwissProt.sequence, PDB.sequence)” expresses the fact that polypeptide_chain is both a sequence in Swiss-Prot or in PDB. For the databases cited above we could consider the classification for protein as follows, where the first field is the relation and the other ones are related attributes: protein (has, name, polypeptide_chain, function) polypeptide_chain(is_composed, SwissProt.sequence, PDB.sequence); PDB.sequence(has, PDB.3Dstructure); Cath.code(has, Cath.domain_def); PDB.3Dstructure(is_composed, Cath.domain_def) SwissProt.sequence(has,SwissProt.description, SwissProt.keywords); protein.function (is_composed, SwissProt.keywords);
The Mediator accepts requests from the user and retrieves the information if the searched data (exploring the DSO) are semantically correlated. It is worth noting here that the Mediator should implement a logic having a definition of query with different abstraction levels (initially, we planned to use the SQL standard language but now we are considering other hypotheses, providing a request virtualisation layer). The Mediator engine coordinates the temporal activities of all of the components selecting those available on some nodes of a Computational Grid.
The Mediator component provides some methods,
through a Web Services interface. The Web service server
has been implemented in C, exploiting the gSOAP
Toolkit [
Moreover, we are finishing the Wrapper APIs for the
data banks cited above, to provide a set of primitives to
get access to and interact transparently with different data
sources. Finally, for high throughput applications we are
investigating an approach based on our mechanism called
SplitQuery which provides an efficient fragmentation of
the biological data set and a protocol for retrieving the
fragment, as described in [
Currently, we are exploiting the Globus Toolkit 3.2
pre-OGSI [
Recently, many workflow languages have been
defined such as Web Services Flow Language (WSFL)
[
As an application of ProGenGrid, we present a workflow modelling the process of searching similarity matching among proteins. Figure 6 shows an activity diagram specification of the similarity search process. This process starts by supplying a target protein < IDProtein > or its FASTA format (in this example, the protein target is 1LYN), the search procedure accesses the database and all of the information about target protein is recovered from the Swiss-Prot database.
To date, we are using the SQL language like that for our experiment. In particular, given the input protein X (1LYN), and indicating with Yi , i ∈ (1, … 200000) a set of sequences extracted from Swiss-Prot, the following query first selects all sequences from Swiss-Prot whose alignment score is greater than a threshold value score, and then, using the sequence Accession Number, it selects from PDB the structural information related to such sequences: select Y.Structure from PDB where Y.AC in (select Y.AC from Swiss-Prot where align[blastP(X, Yi)] > score)
The large amount of data sets that today is available from geographically distributed storage sources, is making data integration increasingly important. Integration of data demands significant advances in middleware; distributed infrastructures such as Grids and Web Services can be used for data integration.
In particular coupling these with ontologies is a promising approach to model bioinformatics sources. In this paper we presented the architecture of a semanticsenriched Data Access and Integration service for biological databases. The proposed system extends the classical mediator approach in data integration by introducing domain ontologies in description of data sources and exposing services through the Web Services approach. Compared to other approaches, our system uses Grid protocols such as GridFTP and GSI for fast and secure exchange of data.
In our architecture wrappers are created manually and added to the mediator modifying its source code. We are now focusing our efforts to build a dynamic mediator through semantic mediation. It will allow using semantic information about data sources, such as query capabilities, data provenance, data scheme, etc. The main goal is to provide a method to add wrappers without source code modifications. A secondary goal is a tool for automatic wrapper generation.
Future work will regard the full implementation of the system and its use inside ProGenGrid, a grid-based service oriented to software environment for bioinformatics applications.