The MIR@NT@N application is an open-access web interface, which can be accessed as a standalone module or through the workflow of M@IA, an environment dedicated to integrative biology analyses 26. MIR@NT@N is built in the PHP programming language for database generation, data integration, analysis scripts (including graph construction and FBL and FFL detection) and interface. It also uses applications included in M@IA: R language (http://www.r-project.org) for statistical computing and Graphviz tool (http://www.graphviz.org) for interaction graph generation. Data can be further processed using any other module of M@IA, such as automatic gene annotation and data mining based on ontology or metabolic/signaling pathways. The MIR@NT@N application is connected to a MySQL relational database integrating information on biological entities and their regulations and interactions. The MIR@NT@N help main page includes an overall description of each section, including a quick tutorial and example files. To guide users in their analysis, MIR@NT@N also provides a contextual help available within each section, by explaining parameters, checking loaded data, and suggesting analysis refinement.

Biological entities were identified and annotated in the MIR@NT@N database using three source databases: PAZAR (Release 2010, January 2010), which includes JASPAR (Release 2009, October 2009) 182728, for TF annotations, including TFBS profiles as Position Frequency Matrices (PFMs); miRBase (Release 14, April 2010) 25 for the miRNA annotations including localization on genome sequence; and Ensembl (Release 56, October 2009) 29 for TF and coding-gene identifier mapping.

To integrate TF regulations on coding-genes (TF→Gene), we combined PAZAR 28 which provides public TF regulatory data, and oPOSSUM (Release 2.0, January 2007) 19, a large scale database which among other features, predicts TFBSs conserved between species, using TFBS profiles from the JASPAR database. Further, we extracted from oPOSSUM all TF→Gene regulations predicted in the 10 kb upstream and 5 kb downstream region of genes, with a score threshold of 0.85, and a high conservation level (top percentile of 0.010 and minimum identity of 80%). For each of the JASPAR profiles, we calculated the correspondence of the scores with empirically derived p-values for a common reference DNA sequence (see "Motif Scoring Procedure and Computation of JASPAR Profile Matrix Score p-values" section on MIR@NT@N website for more details) and established that, for 97% (127 of 130) of the binding site profiles, the applied 0.85 threshold corresponds to a p-value no more permissive than p < 0.01. Present databases 212223 do not provide sufficient information about TFBSs within genes encoding miRNAs (TF→miRNA) required for building a large-scale meta-regulation model. TransmiR provides a limited number of experimentally validated regulations for multiple species 2022. MiRGen offers the downloading of large-scale predicted regulations, but only for Human and Mouse, and without TFBS scores and locations 21, whereas PuTmiR provides scores only for Human 23. Regulation of transcription of coding and miRNA genes has been proposed to be similar. This is based on the observation that promoter regions of both share common features such as the presence of CpG islands and specific histone modification markers 30. In further support of common regulatory mechanisms, it has been shown that a same transcription factor can regulate both, protein-encoding and miRNA genes 31. Thus, we have used a standard TFBS detection algorithm 3 and TFBS profiles from the JASPAR database to predict TF→miRNA regulations on a large scale. PFMs were converted into Position Weight Matrices (PWMs) and used to predict potential TFBSs in 10 kb sequences located upstream of miRNA precursors, extracted from Ensembl database, according to pre-miRNA localization provided by miRBase. To limit the noise of false predictions, only predicted TFBSs with a score higher than 0.65 were integrated into MIR@NT@N database.

To refine the TFBS prediction on the miRNA upstream sequences, we provide additional information on TFBS location within "CpG islands" (CGI), regions which are frequently associated with promoter regions 3032. CGI were predicted (for Human, Mouse and Rat) with CpGcluster 33, a distance-based CGI-finder algorithm, and CpGProd 34, a tool that identifies promoter regions associated with CGI.

To integrate miRNA-dependent regulations (miRNA→gene) into MIR@NT@N database, we combined the miRBase Targets database, rebranded as MicroCosm (Release 5, September 2009) and hosted at the EBI (release 5), and microRNA.org (Release September 2008) 35. Each resource can be used, through the MIR@NT@N application, separately with scores (from the minimum score of 13 to maximum score of 23 for MicroCosm, and from the minimum score of 140 to the maximum score of 205 for microRNA.org) derived from the miRanda algorithm (John et al., 2004), or simultaneously with a unified score (derived by a non-linear transformation and distributed uniformly between 0 and 1).

In addition, we integrated protein-protein interactions from the UniHI database 36, motivated by the idea that clustered miRNAs can coordinately regulate protein-protein interaction networks 37.

Thereby, for 7 species (Caenorhabditis elegans, Danio rerio, Drosophila melanogaster, Gallus gallus, Homo sapiens, Mus musculus and Rattus norvegicus), MIR@NT@N database contains 3 638 miRNAs, 335 TFs, 68 202 coding-genes as well as a large number of predicted interactions for a common standard score threshold of 0.85 (211 783 miRNA→Gene, 32 224 TF→miRNA and 273 264 TF→Gene).

The MIR@NT@N database is publically available on the website, which proposes 1) a dump file of the database in a SQL format, 2) a file (tab-delimiter format) with all TFBS scores calculated from miRNA upstream sequences and TF profiles from PAZAR, and 3) a file (tab-delimiter format) of the meta-regulation network, combining all regulations (TF→miRNA, miRNA→Gene and TF→Gene) for a common standard score threshold of 0.85.

The MIR@NT@N application works within a meta-regulation network model (Figure 1) in order to a) identify novel major regulators and targets based on an input list of actors, through interaction graph analysis and sub-network detection; and b) construct networks with well-defined actors with a presumed role in a given context.

Thus, two types of queries are involved. The first type allows searching for novel key actors in a biological context, using TF/gene/miRNA lists as input (including quantitative expression profiles generated by transcriptomics/proteomics experiments). This query includes three sections: (i) "Transcription Factor regulation" which statistically predicts potential TFs regulating a list of miRNAs, or conversely miRNAs regulated by a list of TFs; (ii) "miRNA regulation" which statistically predicts the significant targets of a list of miRNAs or the miRNAs targeting a list of genes; and (iii) "Regulation Network" which combines both TF and miRNA regulation predictions to reconstitute meta-regulation networks and allows detection of regulatory motifs such as FBL or FFL. The second type of query provides an overview on any TF, gene or miRNA, including their interactions: The "Quick Search" rapidly retrieves information on any actor, its regulators and/or targets, while the "Quick Network" generates regulation networks from a list of actors presumed to be involved in a particular biological context, and also allows the extraction of sub-networks including regulatory motifs.

As described below, the performance of MIR@NT@N was evaluated with published, experimentally validated data and further highlighted in a biological case study on epithelium to mesenchyme transition (EMT). EMT is an evolutionary conserved biological process involving the reprogramming of regulatory networks, including TFs, miRNAs and their targets, in epithelial cells during gastrulation, neural crest cell migration in embryogenesis. In adults EMT is reactivated in pathological situations such as wound healing, carcinoma progression, and fibrosis 143839.

This section reports potential TF→miRNA regulation given a list of TFs or miRNAs to identify novel TF regulators and miRNA targets. The result is a table of TFs or miRNAs, filtered and ranked by their relevance according to several criteria (Figure 2A) including: the quality score, the Fisher test p-value, the number of TFBSs and TF→miRNA regulations. Results can be visualized through an interaction graph (Figure 2B) using a gray scale canonical color code to convey prediction scores. To facilitate the detection of regulatory clusters, the graph includes expression information (if provided as input) using the green/red canonical color code. In addition, to identify clusters of miRNAs regulated by the same TFs, or clusters of TFs which regulate the same miRNAs, an analysis of the interaction graph provides "square" and "curvature" graphs 26. To refine the prediction analysis, all corresponding TFBSs can be visualized through a user-friendly interface (Figure 2C) which provides the binding sequence, its length, the quality score, the localization on the miRNA upstream sequence and in the genome (with a link to Ensembl), and information about predicted promoters using a canonical color code for prediction scores (from yellow to red). All results (tables and graphs) provide links to external knowledge sources (PAZAR for TFs, miRBase for miRNAs, Ensembl for genes and TFBS localization). Results can be exported and stored for further analysis, using for instance the M@IA environment 26 or external applications.

To illustrate the performances of these functions, we identified TFs predicted to regulate the miR-200 family, including miR-200a, miR-200b and miR-429, which are important for the maintenance of the epithelial phenotype and in the prevention of EMT 40. Using stringent criteria we identified four TFs (Figure 2AB) including ZEB2 which has recently been reported to directly interact with E-boxes of the miR-200 promoter 14. The predicted TFBSs of ZEB2 can be located on the miR-200 promoter by clicking on the ZEB2 table, yielding 1 to 9 sites with the criteria 0.9 and 0.65, respectively (Figure 2C). Interestingly, one of the predicted TFBSs is located within the experimentally identified region of the miR-200 promoter (Bracken et al., 2008) shown to be negatively regulated by the related transcription factor ZEB1, mediated through paired E-boxes.

This section determines potential miRNA→Gene regulations from a list of miRNAs or other genes to identify novel actors, i.e. miRNA regulators and targeted genes. The result is a table of genes filtered and ranked by their relevance to the input list of miRNA, using alternative criteria (Figure 3A): MicroCosm and/or microRNA.org scores (or corresponding unified score), Fisher test p-values, number of targets per miRNA and the number of targeted sequences (boxes) per gene. Inversely, this section can provide, using the same parameters, a list of miRNAs predicted to regulate a given list of genes.

As described above in the "TF regulation" section, results can also be filtered using a specific list of data to contextualize the study, visualized through the same type of interaction graphs (Figure 3B). We illustrated this feature by predicting genes that are potentially targeted by three miR-200 family members. MIR@NT@N predicted 934 genes to be at least targeted by two miR-200 family members, using the criteria described in legend of Figure 3. As these miRNAs are known to be down-regulated in EMT 40, we contextualized the study with a biological filter using a list of 132 genes found to be up-regulated in experimentally induced EMT 1741, reasoning that messengers with negatively correlated expression levels may be targets of the miR-200 family 1741. Twenty genes were predicted to be targeted by miR-200 family members (Figure 3A and 3B). The list included FN1, an experimentally validated target of miR-200 42, genes reported to play an important role in EMT 4344 as well as genes with so far no described role in this process, yielding valuable hypotheses for experimental investigations.

This section combines "TF regulation" and "miRNA regulation" interfaces to allow the construction of meta-regulation networks (Figure 4A), with an orientation towards the detection of network motifs and the identification of multiple target genes, for both TFs and miRNAs. Within a specific context, the user may identify, from a list of miRNAs, both novel molecular actors and the nature of the regulation, highlighting fundamental regulatory motifs 10. These motifs include FBLs consisting in a reciprocal regulation of a TF and a miRNA (Figure 4B), the TF controlling the miRNA and the miRNA regulating the TF 45. The FBL modulates the activity of regulators, which is crucial for the spatio-temporal control of their function. On the other hand, a FFL is a regulatory system in which a regulator A regulates another regulator B, and both regulators regulate a common target C 101146. In MIR@NT@N, FFLs can involve a miRNA regulator (FFL-miRNA, Figure 4C) or a TF regulator (FFL-TF, Figure 4D). In addition, MIR@NT@N includes the concept of indirect FFLs (Figure 4E) in which the regulation of the miRNA by the TF is exerted by an intermediate TF.

Users can inform the system by providing a list of molecular interactions. For example, the user can use a list of miRNA→gene interactions experimentally inferred from microarray data combining genes and miRNA expression or a list of published TF→miRNA interactions. For this purpose, published and experimentally validated TF→miRNA interactions 202223 are provided and can be used as a filter.

To demonstrate regulatory motif detection, we analyzed TF→miRNA regulations from published data by Qui et al., including TransmiR data 2022. For the 19 human TFs found in common within the MIR@NT@N and Qiu databases, we observed that 81% of the interactions listed in the Qiu database were predicted by MIR@NT@N with a TFBS score higher than 0.65, and 43% with a TFBS score higher than 0.85 (Additional file 1). Using entire MIR@NT@N database, we extracted putative FBLs (Figure 5A) and FFLs (Figure 5B), including well-documented FBLs implicating an E2F TF family member and several miRNA families 4748, the ZEB "Zinc Finger E-Box" TF family and the miR-200 family 4950, and YY1 and hsa-mir-29a 51. Moreover, we predicted hsa-mir-29a to be regulated by NFKB1 and MYC (Figure 5B), consistent with previous reports 5152 and only recently identified to be co-regulators of their common target mir-29a 53.

Collectively, these results underline the efficiency of MIR@NT@N to generate an overview of a regulatory network and to detect core sub-networks within a biological system.

Quick search and Quick network interfaces allow searching for regulations between known or assumed actors of a biological context. The "Quick Search" section is a full text search engine that provides data pertinent to specific entities (miRNA, TF or target gene). Information about each biological item is available through hypertext links to external data sources (Ensembl for genes, PAZAR for TFs and miRBase for miRNAs). Potential TF/miRNA regulations and predicted TFBSs are accessible through an internal MIR@NT@N application pipeline.

The "Quick network" is a powerful application to extract information from a list of TFs, miRNAs and other genes with a presumed function within a biological context, as supported by literature or experimental data. The user can retrieve corresponding regulatory predictions and generate a network of predicted interactions as a comprehensive graph, yielding information on the interaction mechanisms of the analyzed actors. Functional motifs (FFL and FBL) can be detected to identify major actors and targets organized into regulatory sub-networks. The respective quality score thresholds of TF and miRNA regulations can be modulated through a cursor and information about protein interactions can be integrated into the network (described as experimentally validated in UniHI). A cross-species network analysis is possible by selecting different species associated with the input symbol (e.g. the input symbol "hsa-mir-200a" will be changed into "mmu-mir-200a" if the "Mus musculus" species is selected). The output is an exportable interaction graph recapitulating all predicted interactions and which is linked to external resources (Figure 6).

The miR-200 family served as an example to illustrate how the "Quick Network" interface generates regulatory networks in a given context (Figure 6). The generated network recapitulates the results described above and integrates the predictions of the TFBS on coding genes (Figure 6A). The FBL function suggests the presence of a double negative FBL between mir-200a, mir-200b, mir-429 and ZEB2 (Figure 6B), as is described in the literature 54. Target genes already described in EMT, such as ROCK2 44 and TEAD1 55 were highlighted from the FFL network (Figure 6C).

MIR@NT@N, which takes advantage of the M@IA environment 26, can be readily extended to include additional miRNA target prediction databases (such as TargetScan 56 or PicTar 57) or more TF binding profiles from collections that use a standard PFM format. The PWM methods utilized within MIR@NT@N are well-established, but likely to be replaced with more advanced models in the near future. High-throughput sequencing coupled to chromatin immunoprecipitation now routinely generates collections of ~103 binding sites, providing richer descriptions of binding properties of TFs. New algorithms are emerging which build on such data to describe patterns using higher-order models to account for interactive effects between positions. However, the rapidly emerging changes have not stabilized, so we applied the established methodology within the source database in the oPOSSUM system. We intend to upgrade MIR@NT@N when a new motif scoring procedure is supported by the JASPAR database of binding profiles.

Moving forward, novel data classes will be implemented into MIR@NT@N, such as histone modifications or alternative splicing that play central roles in gene expression and for which databases are already available 5859. We will incorporate more knowledge sources, such as known promoter sequences and experimentally validated TF-miRNA regulations 202260.