Nucleic Acids Research

Computational infrastructure

RegulonDB uses a relational database scheme. The design of the database is explained in detail in ref. 1. We have migrated the database into Unix using Sybase SQL-Server, a relational database manager system. The relational design has been kept as previously described, with the following tables: operons, promoters, genes, regulatory interactions, polypeptides and proteins, signals and conditions, and finally references.

Scripts were written in Transact-SQL language and Sybperl (http://www.mbay.net/~mpeppler/ ) in order to make the database available on the web. To generate graphical web interfaces, we used Perl (2), java applets and JDBC-driver. In addition, Reference Manager (Research Information System, Inc., Camino Corporate) has been used to organize a parallel literature database. Biological information has been gathered from the literature using Medline and PubMed (3; http://www.ncbi.nlm.nih.gov/Entrez/medline.html ), as well as other biological database sources as mentioned below.

The web interface for RegulonDB can be found at http://www.cifn.unam.mx/Computational_Biology/regulondb/ . The menu on the left of the main page offers three types of search procedures, what we call the simple search, the search for regulatory mechanisms, and the search for operon organization. These were built keeping in mind the main biological content of the database centered, as mentioned before, in mechanisms of regulation, and operon and regulon organization. Furthermore, the results of each search can be displayed graphically. These graphics are built with the feature-map program described in ref. 4 (http://copan.cifn.unam.mx/Computational_Biology/yeast-tools/ ). Browsing by operon organization, users can get a graphical automatic display of the internal organization of operons, with the genes, promoters and sites indicated, with symbols in different lines associated to their direction in the chromosome.

Browsing by regulatory mechanism permits, for instance, to search for the complete set of regulatory sites of a given protein, its alignment and weight matrix, and to see such sites in a single graph in relation to the transcription initiation of each regulated promoter. Alternatively, if the name of the promoter is indicated, all sites that regulate it can be obtained. A `help' button permits the user to go through the documentation explaining in detail the options implicit in such searching procedures, and the restrictions in the syntax for inputs in these browsing tools. The help pages include examples in order to guide new users through the basic functionality of the search forms.

All bibliographic references are now described uniformily by means of active MEDLINE numbers used as direct links to PubMed. In this way the users can go directly, for instance, from a regulatory site to the reference in Medline reporting the experimental work that supports it.

Finally, we have also developed tools for the internal management, updating and correction of the database. These include programs that check for the consistency of new data with exisiting data, ensuring data integrity. We verify, for instance, that promoters are well located upstream of the beginning of the transcribed gene; that two distinct entries do not have the same name; that genes within an operon are all transcribed in the same direction; and that one gene belongs to one and only one operon. Note that this last restriction is a consequence of the structural definition of operons implemented in RegulonDB. Complex operons can transcribe subsets of genes under different physiological conditions by means of different promoters and internal terminators. The design of RegulonDB (i.e., adding a termination feature to the genes table) can describe such complex operons, as well as the absolute position of sites, promoters, and individual genes in the genome sequence of E.coli.

Changes to the biological content

Information about the clustering of genes into operons was initially obtained from Mary Berlyn (E.coli Genetic Stock Center: http://cgsc.biology.yale.edu/ ). Based on that initial input, we searched in the literature and started correcting a large number of operons. One of the reasons that explain why this information required extensive verification is that biologists use the terms of `operon' or `transcripton unit' with different precise meanings. RegulonDB is based on the definition of an operon as a set of co-transcribed genes and their associated regulatory elements. This permits a precise notation for names of operons. For instance, if three genes such as araB, araA and araD form an operon, the operon is named araBAD. In the more complicated case of an operon formed by genes with unrelated names, we define the name of the operon concatenating the gene names. For instance, the operon containing aroF and tyrA is named as the aroF-tyrA operon; and that containing aroL, yaiA, and aroM, is named the aroL-yaiA-aroM operon.

Additional corrections come from the multiple names given to individual genes, and to the positioning of genes in the complete E.coli K-12 sequence. A brief summary of such changes is shown in Table 1. Our absolute coordinates are based on the current, so-called m52 version, of the genome described in ref. 5. New additions to the database include, among others, the identifiers for genes used in the Blattner laboratory, or `b-numbers'; and the set of predicted operons, promoters and regulatory sites. A more detailed description of the method used to predict regulatory sites is contained in ref. 6. Briefly, the strategy followed for these predictions was to complement what is already known from experiments reported in the literature. Thus, promoters were searched and predicted only in those upstream regulatory regions that do not contain a single promoter experimentally described. Similarly, genes were clustered in operons by means of an algorithm only in cases where we have no experimental information of operon organization. In this way, predictions complement the experimental knowledge, generating together an organized structural description of the complete genome into sets of operons. RegulonDB will then contain this mixture of known and predicted objects, which will be clearly distinguised by a `type' field within operons, promoters or regulatory interactions. It is important to emphasize that the set of predictions may change with time, as new improved methods will be utilized, and as we continue gathering information from the literature.

Table 1. Summary of corrections

Number of promoter corrections
Position	Name	Sigma factor	Strand
106	15	2	3
Number of `regulatory interaction' corrections
Type of evidence	Medline accession number
20	74
Number of gene corrections
Name	Synonyms	Strand
7	6	12

AVAILABILITY

RegulonDB 2.0 can be accessed though the URL: http://www.cifn.unam.mx/Computational_Biology/regulondb/ . We kindly ask users of RegulonDB (2.0) to cite this article.

ACKNOWLEDGEMENTS

This work was supported by grants from DGAPA and Conacyt to JC-V. We want to thank the availability of the operon information to Mary Berlyn, as well as to Monica Riley for her classification of gene functions, and function of gene product information (together with Guy Plunkett III). We thank the availability of the E.coli sequence to the group of Fred Blattner. We also appreciate useful discussions with Araceli Huerta, Peter Karp and Edgar Wingender.