Transcriptome surveying using microarrays has become an established and widespread technique1. Although glass based microarrays and oligonucleotide chips are common in the gene expression profiling (GEP) landscape, Nylon supported microarrays coupled with radioactive detection, either home made 23456789 or from industrial suppliers 1011121314, are an alternative still favored by some researchers15, including the NIH's National Institute on Aging 16171819. The resilience of this platform can be explained by the easy setup of this technical combination, its high sensitivity achieved without target amplification20, and its cost effectiveness16. Technological development for this platform is ongoing, as demonstrated by a two 'isotope' dual channel variant that uses real time emission integration21.

Regular increase in microarray reporter densities together with falls in unitary costs have meant experiments routinely generate tens of millions of data pieces to store, search and analyze22. For all but occasional microarray users, dedicated laboratory information management systems (LIMS) are now a requirement. Arguably one reason bench scientists are yielding to laboratory databases is increasing pressure from journal editors to have data appropriately submitted to international repositories as a prerequisite for publication23.

Amongst the gene expression profiling LIMS that have been reported 2425262728293031323334, glass dual channel Cy5/Cy3 and oligonucleotide chips are extremely well catered for. However, Nylon based microarrays are at best marginally supported by these software, with more substantial Nylon support only to be found in commercial products35. Drawing on experience with a previous LIMS36, the specific functionalities that we have found critical for comprehensive Nylon based microarrays modeling include the ability to record so called 'vector' hybridizations (quantitations of spotted reporter amounts), the stripping and recycling of microarrays37 and to model controls spiked into RNA samples 2038. Finally, the pharmaceutical industry manufacturing and regulatory rigor required for the development of diagnostic applications are typically not of major concern in freely available LIMS.

Here we report the development of MAF (MicroArray Facility), a LIMS designed to accommodate the following desiderata:

-stringent quality control, traceability and audit tracking to meet industrial requirements

-multi-platform support (i.e. Nylon, glass, and oligo-based microarrays)

-rich data annotation for MIAME standard compatibility39

-support for oncogenomic projects (i.e. clinical data)

-dynamic cDNA reporter annotation using public data banks (notably UNIGENE40)

-MAGE-ML enabled41

-import robotic equipment files in native formats to reduce error-prone reformatting

-multi-user privilege environment to promote data sharing and ensure confidentiality.

The scope of MAF (as defined in 42) is a local LIMS serving a community of researchers, such as found around academic shared facilities or commercial array and service providers. The MAF user interface is entirely web browser contained, thus there is no technical contra-indication to using MAF as a means to publish datasets over the Internet. However one-stop shops for gene expression data have innumerable advantages for public data mining43; thus we strongly recommend uploading public data to international archives such as ArrayExpress44, CIBEX45 or GEO22 as a mechanism for publishing data.

MAF is a LIMS sensu stricto in that it records, tracks, structures, searches and reports all information required to establish gene expression profiles. High level downstream data analyses are carried out by exporting selected data to any of the myriad of dedicated analysis packages such as Cluster46, BioConductor47, MeV26 or ProfileSoftware48.

MAF follows a client-server architecture implemented as a web-based application, allowing simultaneous multi-user access to a central database. Client browsers connect to an Apache server in a Unix/Linux environment. The application is entirely written in Perl; since the Perl packages rely on the abstract DBI module, switching between different Relational Database Management systems (RDBMS) is as simple as changing a single line in the MAF configuration file. MAF has been implemented, tested and validated for the Oracle 8i and the PostgreSQL RDBMS. With Oracle 8i, certain Oracle-specific features such as backup tools and transportable table spaces can also be used. Having deliberately avoided RDBMS specific SQL syntax, we believe that MAF is easily portable to other SQL database platforms that have a Perl DBD driver.

The relational database underlying MAF – composed of 2258 fields held in 215 tables – is called ELOGE. Its schema extends the conceptual ArrayExpress design44 which integrates microarray design and manufacturing, sample description, hybridization and data acquisition. Through a 5-year development cycle, ELOGE was considerably expanded to integrate fine grain modelling of wet lab routine procedures such as plate management, PCR quality control annotations, sequencing results, analytic validation, and GLP (Good Laboratory Practices) compliant protocols. The database scheme is designed to avoid computation-intensive queries and optimize user interface responsiveness.

MAF is accompanied by two complementary software modules (Fig. 1): the Gene Finder (GF) and Clone Chooser (CC) which respectively manage gene and clone lists, as well as provide effective mapping from one to the other. Genomic data relevant to all three modules are imported into ELOGE every two months from several commonly used public databases (GENBANK EST, UNIGENE, SWISSPROT, ENTREZ GENE, GO and REFSEQ), allowing automatic and thorough annotation of genes and cDNA clones. Together, the MAF, GF and CC modules constitute a package called Discovery Software.

MAF user data are collected through web form cascades (Fig. 2). Where appropriate (e.g. array layouts or image quantitations) data files are uploaded using background queue processing to avoid tying-up the interface. Many instrument and third party software data files are thus directly imported using a large set of data formatting drivers (Table 1). MAF web forms can be used to add, update or view data. However, browsing MAF data is best achieved through a collection of hyper linked data reports specific for each step of GEP processing (Table 2).

Confidentiality is ensured through user login/password authentication and project containers. Access to a project's objects (arrays, plates, hybridizations, quantitations etc.) is controlled by the project owner through read, write, update, and delete independent privileges. Unique transaction ID's for each request, inactivity time-outs, enforced single user sessions and SSL network encryption complete the MAF security strategy.

We have developed MicroArray Facility, a software tool for the management of microarrays which offers extended Nylon functionalities not found in other freely available LIMS (Table 3). All gene expression profiling steps from cDNA clone management to spot measurements are represented in the MAF database with annotation granularity compatible with MIAME.

Importantly, the MAF system has been tried and tested in both academic shared facilities and industrial environments managing cDNA and Affymetrix gene expression projects. Running in production for five years, MAF has established itself as a central information hub in the laboratory. Investment in data entry is rewarded by providing researchers with fast answers to common queries (e.g. "what is the expression profile of this new marker in our previously tested tumors?") and by helping extract more biological meaning from collected data.

• Project name: MicroArray Facility (MAF)

• Project home page: http://tagc.univ-mrs.fr/bioinformatics/maf/

• Operating system(s): Platform independent

• Programming language: Perl, SQL

• Other requirements: Web Server (e.g. Apache), RDBMS (Oracle or PostgreSQL), MAGEstk

• Licensing: Royalty free, non-exclusive, non-transferable INSERM license for academics (excluding SOPs)

• Any restrictions to use by non-academics: Discovery Software license from IPSOGEN SAS

PHi wrote the first draft of the manuscript. PHi, SDeb & SG conceived of the study, and participated in its design and coordination. PHo developed the MAF module. RT & SDer developed the CC and GF modules. EB and JR participated in MIAME/MAGE-ML implementation. All authors read and approved the final manuscript.