In order to import microarray databases into SRS (version 7.1.3), an Auto-Upload Tool was built (Figure 1). This PHP-written tool allows one to store databases of a predefined format into a user-owned and password protected directory on a local SRS server [see Additional file 1] 29. In this directory, databases can be managed, viewed and uploaded into SRS. In the "edit-database" interface, links to other databases can be specified for each field. A standard view can be generated and the location of the dataset in the SRS directory determined. Finally, permissions can be set to control access to the various datasets in SRS. Upon uploading, the Auto-Upload Tool will generate the files required for SRS: (i) the SRS data file in which the spreadsheet input file is converted into a flat-file database, (ii) an Icarus syntax file (.is-file) which describes the layout of the flat-file database, (iii) a database index file (.i-file) which describes the way in which the different fields need to be indexed for the SRS system, (iv) a database view file (.view-file) in which a standard view is defined, and (v) an information file (.it-file) which can harbour a description of the dataset. These files are automatically placed into the SRS directory after which the Auto-Upload Tool updates the srsdb.i, user.i and site.i files. These files describe the name of the database, where the files are located (srsdb.i), user permissions (user.i), and configuration of the different database groups (site.i). The srssection command within the Auto-Upload Tool implements the changes in the configuration files after which srscheck and srsdo perform indexing of new databases and set links. Incorporation of a new dataset using this tool generally takes place within minutes. On our local SRS server, a DQS (Distributed Queuing System) batch-queue is installed to prevent data loss or corruption of datasets in case multiple users are editing datasets at the same time.

An important feature of storing and linking microarray data in SRS is the accessibility of the datasets for other programs. As an example, we generated a PHP web interface for the Venn Mapper program that retrieves microarray data from SRS to calculate the statistical significance of the number of co-occurring differentially expressed genes in any combination of two experiments 26. The functionality of the original Venn Mapper was enhanced by enabling the use of different ratio cut-offs for different microarray experiments. Upon login, the interface displays all microarray databases indexed in SRS to which the user has access. After selection of the datasets a second screen shows all fields (such as individual array experiments or averaged group ratios) of the selected databases. The microarray experiments of interest can be selected for Venn Mapper analysis after which the requested data is linked and exported from SRS into the Venn Mapper program. The output of the program is available for viewing and downloading. Information requests from the interface to SRS are made through the SRS getz command 30. This powerful feature of SRS makes the integrated databases accessible to any external program.

With respect to the microarray database set-up, there are two important considerations. First, in our experience, microarray data mining often starts with selecting genes based on their differential expression. Differential gene expression is best determined using statistical evaluation of the data based on sufficient technical and biological replicates 2728. Dependent on the statistical test and microarray platform, raw gene expression data and/or ratio calculations can be utilized. Second, making changes to a microarray dataset in SRS is impractical and datasets should be fully built before they are imported. This means that raw expression data and ratio calculations should be normalised and flagged and represented in a common format (such as log2). Importantly, statistical evaluation should be included. For simplicity in representation, datasets can be summarised in, for example, an average of all "normal" and "cancer" samples and additional fields of log2 "normal/cancer" can be included.

Linking of databases should be based on invariable and unique indexes. Links based on identifiers such as UniGene cluster identifiers that are regularly re-assigned, forces one to repeatedly update all databases that include such a denominator. Invariable links based on DNA sequence assignments such as GenBank and ProbeSet identifiers (IDs) are therefore more appropriate linking indexes. We recommend including only one of those invariable and unique linking fields in the microarray dataset to avoid the need for a regular update. Since many biological databases do not use these hard links, connecting microarray datasets to other databases can be achieved through a coupling file and/or by making use of the web of links provided by the biological databases (Figure 2). A coupling file can be minimal, containing for example only a RefSeq ID with the appropriate array spot number, or can contain a variety of indexes such as GenBank, RefSeq, SwissProt, OMIM, LocusLink, KEGG, GO, GeneCards, UniGene identifiers, that directly link to the different databases. Since some of these links are variable and biological databases keep growing with information, coupling files must be regularly renewed to update the various links.

Instead of or in combination with a coupling file, one can make use of the links provided in the various biological databases (Figure 2). For example, the GenBank accession code in the microarray dataset can be linked to the UniGene database. The LocusLink database is linked to the UniGene database through RefSeq accession numbers and also contains IDs that for example link to EMBL, OMIM, and SwissProt. In this way, almost all biological databases are linked through a network of direct and indirect connections. In case multiple roads lead to the same databases, SRS utilizes only one route. By assigning values to each link, the route taken is the one having the lowest sum of link values, even when this results in a lower number of connected fields (genes). Although SRS can be forced to take a specified path, one should be careful to be dependent on many different databases. Inconsistencies in and incompleteness of databases are accumulated when linking occurs in sequence.

Linking to the network of biological databases through a coupling file has various advantages. One can establish directly validated links to each database, including databases outside the network. In addition, errors in links can easily be corrected. Platform-specific or overall coupling files can be retrieved from the Affymetrix website 31 and from sources such as Resourcerer 32, KARMA 33, GeneHopper 34, ProbeMatchDB 35, DAVID 36, EnsMart 37, and Source 38. Using these resources, microarray datasets from different platforms can be linked. This includes connecting cross-species datasets using ortholog converters such as HOMGL 39 and HOMOLOGENE 40.

A high accuracy and comprehensiveness of linking are essential for a successful comparison of microarray data from different platforms. The extent of linkage of various databases in SRS was examined (Figure 3). The percentage of fields of scientific databases and microarray datasets that are linked to other databases was assessed. As shown in Figure 2, most microarray databases are directly linked to the scientific database network via a single connection. The U133A and U95 Affymetrix coupling files contain direct links to UniGene, SwissProt, LocusLink, and OMIM. The linkage of the various microarray platforms to UniGene varies between 84% and 96%. On average, 60% of the genes of microarray datasets can be linked to each other via UniGene.

The availability of many scientific databases in SRS, the universality of the system and its free access for academic use, make SRS an excellent mining system for heterogeneous microarray datasets. The Auto-Upload Tool facilitates the exchange of microarray datasets between separate SRS installations. Using a single data file and optional description file, any user can upload the identical data and customize it to their own SRS environment. We would urge researchers and microarray data repositories to make their data available in an SRS format. In addition, microarray software programmers could make their software available in an SRS compatible format or include SRS data export options. The commercial SRS GeneSpring^® Connector and public EMBOSS are examples of such microarray-SRS integration ventures.

We plan to extend our efforts of integrating more microarray databases into SRS. In addition, software tools specific for microarray data analysis, such as Go Mapper and CoPub Mapper will be rewritten for SRS 264142. The CoPub Mapper literature mining program contains databases that store, for each gene, all MEDLINE records mentioning the gene. This directly links microarray expression data to the published literature and allows for co-publication research of gene-gene and gene-keyword combinations.