Import

Common network exchange formats are supported such as SBML [16], BioPAX [17], KGML [18], GML [19], DOT [20], SBGN-ML [21] and SIF [12] thereby enabling the exchange of data throughout the community. Various databases (e.g., KEGG [22]) provide network files which can be imported into Vanted via drag-and-drop. Vanted is directly connected to the MetaCrop and the RIMAS databases. The MetaCrop database [23] contains manually curated information about metabolic pathways of major crop plants and corresponding networks in SBGN [24]. In addition to metabolic pathways the database comprises information about reaction kinetics and gene identifiers as well as related literature references. In order to filter, explore and import this information, the MetaCrop add-on provides seamless access [25]. Besides metabolic networks, gene regulatory networks of the RIMAS web portal [26] can be directly accessed. This information resource comprises SBGN-style networks about regulatory interactions during seed development of Arabidopsis thaliana.

The import of experimental data is preferably done by using XLS templates, which enable a structured import together with meta-data. Alternatively, plain text or CSV files may be used to import large datasets such as gene expression data, but require manual enrichment with meta-data. For unlimited accessibility, persistent storage and exchange of experimental data, the DBE² information system [27] is accessible via the DBE ² add-on. The add-on utilizes ontologies from the Ontology Lookup Service [28] to unify terms such as compound names, species names and measurement units aiming at a facilitated data integration. As Vanted, DBE² supports different data types from numerical data to images, three-dimensional volumes and networks.

Visualization

Networks are represented as graphs composed of nodes and edges with fully customizable visual appearance. Numerous visual attributes such as the position, size, color and frame thickness of nodes as well as the color and thickness of edges and other visual attributes such as labels can be adapted according to individual purposes. In addition, a specialized set of node and edge shapes is provided, which build the basis for an SBGN compliant network visualization. SBGN-ED [29] enables Vanted to adapt networks for all SBGN languages in order to facilitate a standardized visual representation of biological entities. The visualization of such maps can be validated for syntactic and semantic correctness according to the SBGN specification.

Readable network layouts are important to improve the visual representation of networks. Besides the manual layout of network elements, automated graph layout algorithms are provided by calling the external Graphviz layouter API [30] or executing self-implemented layouters based on Tollis et al. [31] such as the force-directed layout, tree layout, circle layout, expression matrix layout, grid layout, subgraph layout and edge-routing algorithms. Further editing or improvement of automatic layouts can be done by manual curation using node merging and splitting algorithms. The latter is important for splitting frequently occurring nodes such as ATP or CO₂ in metabolic networks, thereby preventing edge-crossings throughout the network.

Vanted offers the integration of various datasets into network nodes and edges (data mapping) thereby enabling a network-based view on large-scale datasets. Options for visual representation of experimental data include shape and color coding of nodes and edges as well as more complex visualizations such as bar charts, pie charts, line charts and heat maps. Experimental factors of complex datasets such as time-resolution, varying genotypes and environmental conditions can be represented within one chart. Visualization of charts is performed by calling the JFreeChart library [32]. The FluxMap add-on [33] enables the visual representation of flux data by edge thickness adaptation. This supports the comparative visual analysis of complex flux distributions in an interactive way. Using the HIVE add-on [34] image-based data such as histological cross-sections, microscopy images, photographs and three-dimensional volume data such as NMR and CT data can be displayed in the network context based on a workspace approach and rendered using various 2D-, 3D- and network visualization functions.

Every shape, label, chart and even the selection are realized in Vanted as single Java Swing components placed in the graph window (for further technical details see [35]). Other commonly used libraries such as JUNG[36] render all graphics in a single component. Vanted s approach is harder to implement, but scales better in terms of rendering speed and enables high flexibility in adapting and fine-tuning each component. The highly optimized Cytoscape framework on the other hand scales very good, but does not enable comparable flexibility in terms of visualization of charts, shapes and other graphics.

In general, visualization is the most advanced feature of Vanted. Multiple options and functionalities enable users to generate appropriate visual representations thereby substantially facilitating the gain of knowledge compared to working with data tables. Vanted enables users to interact with up to 10k network elements, but the responsiveness depends on the visual complexity as complex charts, labels and other visualizations as well as high numbers of edge crossings may reduce this numbers considerably down to some thousand elements. For larger graphs, interaction may become unfeasible and algorithms such as automatic layouters consume a considerable amount of time.

Integration

Biological entities such as proteins, genes or metabolites are represented as nodes and any relation between such entities as node-connecting edges (e.g., regulation, interaction or conversion). Both network elements are attributed by technical properties such as visualization parameters (size, position, etc.) and properties related to their biological role. Each network element may contain links to other resources, usually represented as a hyperlink to any web-content such as a database entry. Nodes may link to other networks, enabling navigation and exploration of connected pathways (see also Section Exploration and interaction). Based on the present numerical attributes, for example, size, position and node degree, the user is able to compute new properties such as additional median values, which are stored as new element attributes and may be visualized or exported.

In Vanted, network elements are allowed to have several (alternative) identifiers. These identifiers provide the basis for data mapping which depends on common identifiers in network and experimental data. In case of different identifiers, synonyms have to be defined. For this mapping tables may be used to provide either additional labels for network elements or for biological entities in the experiment data. Mapping tables are simple XLS files, which list the existing names in the first column and additional names in the subsequent columns.

Simulation

Basis of the simulation task is the modeling capability of Vanted. Model reconstruction is based on a given network topology, which is manually created or imported from network files. Subsequently, model attributes such as stoichiometric coefficients, kinetic constants, firing rules and initial markings are added to the network or are already part of the import process (SBML files for example provide most attributes). So far, Vanted does not support the automated reconstruction of networks from external sources as described in [37].

These biological networks are finally transformed into mathematical models in order to analyze dynamic properties and behavioral attributes. The enrichment of metabolic networks with stoichiometric coefficients (represented by edge weights) and the definition of an optimization function is a prerequisite for the constraint-based network analysis. The FBA-SimVis[38] add-on enables Vanted to perform different techniques such as Flux Balance Analysis [39], Flux Variability Analysis [40], Robustness Analysis [41] and Knock-out Analysis. In combination with a dynamic and visual exploration of simulation results, this allows for the comprehensive analysis of metabolism in response to genetic or environmental perturbations. Metabolic networks can also be transformed into Petri nets [42], a second mathematical model, which is used for formal analysis and simulation of biological systems. The Petrinet[43] add-on enables Vanted to semi-automatically transform networks into valid Petri nets, simulate discrete and continuous Petri nets of varying complexity and analyze structural properties. Different visualization and interaction techniques such as brushing can be utilized in order to visually analyze P- and T-invariants, the reachability graph and varying markings of simulation steps.

Exploration and interaction

In terms of exploration of networks and data visualizations, Vanted supports standard interaction methods such as panning, zooming and overview+detail for selected network elements. The editing and rearrangement of network elements as well as the modification of attribute values and calculation of new attributes is possible in an interactive manner. Sophisticated selection and search functionalities provide the ability to find and explore network elements based on attribute values.

Furthermore, recurring entities in large networks or several networks may be linked in order to easily track interconnections between pathways. The GLIEP[44] add-on provides an interactive view for the exploration of interconnected networks by implementing a glyph visualization. Based on these glyphs the user is able to quickly switch between connected networks or to explore the overall interconnectivity using a focus+context technique. Furthermore, the HIVE add-on enables users to collapse networks into single nodes, thereby providing a clear representation of multiple (interconnected) networks. Connections between different networks are retained and link the network-overview nodes, which can be re-arranged or expanded according to user requirements.

On the basis of interaction events such as selection, brushing techniques [45] provide different views on visualized experimental data. The HIVE add-on enables users to explore and compare spatial distributions within a biological system by parallel visualization of segmented images and experimental values in the network view. Hovering over a segment in the image (e.g., corresponding to an organ) results in highlighting the respective measurement values in the network view. Furthermore it is possible to explore large numbers of images in the context of a network. If these images are related to a substance (e.g., GFP reporter expression for genes in a gene regulatory network), the user can integrate the respective images into the network nodes. If a number of nodes is selected, an image matrix is built up, spanning conditions, time points and replicate information. This matrix enables users to compare all images related to the selected nodes and to explore spatial patterns of different substances in the context of a biological network.

Further brushing techniques are provided by the Petrinet add-on for the analysis of Petri net properties such as invariants and the reachability graph. The user can move the mouse over nodes of the reachability graph, triggering the visualization of the respective state in the network visualization view.

Analysis

The analysis of network topology plays an important role for the understanding of interactions between biological entities. Vanted offers to compute several topological properties such as shortest paths between node pairs, network cycles and motifs. The detection of network motifs (such as feed-forward loops) is supported by the possibility to search for user-defined motifs which might be meaningful in the context of certain biological questions. The Vanted add-on Centilib[46] provides algorithms and methods for the computation and investigation of 17 different centralities in biological networks. Such centralities can be used for ranking of network nodes according to given criteria and for the detection of network hubs. Results of the centrality analysis can be explored and analyzed using a brushing-based approach.

The statistical evaluation of experimental datasets is a central part of data analysis. Vanted offers a series of tests for calculation of statistical parameters, for testing the normal distribution of datasets (David Quicktest [47]) and for outlier detection (Grubbs test). For the comparison of measurements with multiple conditions, several t-tests are available such as the unpaired t-test, the Welch-Satterthwaite t-test and the Mann-Whitney U-test with user-defined threshold settings for the calculated p-values. Vanted enables users to perform Pearson’s and Spearman correlation analysis based on the mapped experimental data. Optional settings include a p-value threshold and the number of experiment conditions included in the analysis (see [4] for implementations details).

The calculation of clusters is a frequently used approach to categorize experimental data into functional or behavioral groups. For this task, Vanted supports self-organizing maps (SOM) [14]. A SOM is an artificial neural network, which is capable for the automated recognition of patterns within measurements and is well-suited for the categorization of time series data of biological entities. According to a user-defined number of target clusters, the SOM is trained and cluster attributes are automatically assigned to the network nodes. In addition such assignments can be done manually. The cluster sub-networks may then be independently laid out or colorized in order to visually catch clustered elements at a glance.

For gene expression data Vanted supports the computation and visualization of enrichments in the context of the GO [48] and the KEGG pathway [22] hierarchies. For example, for KEGG the procedure highlights classes of KEGG pathways in which the experimental data enriches significantly by assigning pie charts [49, 50].

Export

Vanted provides a variety of file formats for data storage, publication and exchange. The GML and GraphML file formats are Vanted s native formats and accordingly support the storage of networks together with all related attributes such as layout information and the full set of mapped and integrated experimental data including the visualization options for mapped data. Additional information can be stored and exchanged as new attributes, e. g. a new custom attribute “myAttribute” enables to colorize all nodes with this attribute based on the respective attribute value. Such attributes can be created manually (e. g. cluster information and biological tags) or be the result of a computation (see [35] for further details).

For the exchange of data within the systems biology community, support for file formats such as DAT [51], SBGN-ML (provided by the SBGN-ED add-on) and BioPAX is implemented. Vanted additionally supports the SBML file format which allows for the storage and exchange of stoichiometric and kinetic models. When working with the Petrinet add-on, the Petri net and its configuration can be exchanged using the PNML file format. Experimental data which has been mapped onto a network can be extracted and exported using XLS sheets. The CSV format is supported for different kinds of node attributes as well as the export of analysis results such as correlation coefficients. All data types which are supported by Vanted (numerical data, images, three-dimensional volumes, networks) can be uploaded to the DBE ² system for persistent data storage and exchange. Please note that Vanted usually serves as a data sink and the conversion between different file formats is not in the focus of the tool. Network topology (including labels) on the other hand is preserved in most cases.

Laid out networks can be exported to several graphic file formats, including raster images (PNG, JPG), as well as vector images (SVG, PDF, PPT). These file formats are well suited to be used as images in publications, presentations or as a basis for further graphical editing. Furthermore it is possible to export integrated networks as browseable and clickable images, embedded in HTML web sites. Those images can contain web-links to web resources or public databases. The publishing process of these web sites can be done in a semi-automatic fashion [52].

Import

The import of enzyme activity data, metabolite data and morphological parameters of different Arabidopsis thaliana accessions from climate class A is realized using the Vanted XLS template (see Additional file 2). Experimental data may also be persistently stored in the DBE ² database, enabling file sharing and on-click import of such experimental data into Vanted. In parallel to the import of the experimental data, 38 metabolic reference pathways are loaded from the MetaCrop database and merged into one SBGN network. Subsequently all reference pathways are assigned to their respective cellular location and the pathways in each subcellular compartment are connected to each other by merging identical metabolite nodes. Finally a network layout is performed in order to optimize the edge routing and distance between nodes, resulting in the network which can be found in Additional file 3.

Visualization and integration

During data mapping, experimental data is integrated into the network by the visualization of corresponding charts inside the network nodes. To unify the identifiers in the network and the experimental dataset, a mapping table is used for the enrichment of network nodes with alternative identifiers (Figure 3a and Additional file 3). Subsequently, metabolite data is mapped to the nodes representing metabolites (simple chemical glyph) and enzyme activity data is mapped to nodes representing enzyme nodes (macromolecule glyph). New nodes for morphological parameters are added during the mapping process, as they are part of the experimental data, but do not occur in the network. The mapped experimental data is visually represented by bar charts inside the glyphs resulting in a data-enriched SBGN network (Figure 3b and Additional file 4).

Analysis

In order to identify similarities in the profiles of all accessions of climate class A, 1:n and n:n correlation analyses are performed. In case of the 1:n correlation analysis, the morphological parameter fresh weight (FW) is chosen as the target parameter and correlations were calculated to all other metabolic parameters in the network. Based on the resulting correlation coefficients network nodes are color-coded according to the correlation coefficient r (Figure 3c and Additional file 5). This visual representation of correlation results enables biologists to easily identify metabolic parameters with important influence on plant morphology at a global scale.

For the n:n correlation analysis, all metabolic parameters in the network are correlated with each other, including all metabolite and enzyme activity data as well as the data of morphological parameters. The resulting correlation values are visualized by generating new edges between correlating nodes. These edges are color-coded according to the negative (red) or positive (blue) correlations calculated with p≥0.95 and |r|≥0.6 Pearson’s product-moment correlation. The resulting network is used to generate a correlation network at a pathway level, independent of the order of metabolic reactions within a pathway. Consequently, the metabolic dataset is used to generate new nodes in a network-independent manner which are then categorized according to the metabolic pathway (e.g., Glycolysis, TCA cycle) and laid out as pathway-specific circles (see Figure 4). During the n:n correlation analysis Vanted generates edges between nodes with data profiles of significant similarity thereby giving an overview about intra- and inter-pathway dependencies and allows for drawing conclusions about the interaction between single parameters. For example, the levels of amino acids show strong positive correlations among each other and with levels of TCA cycle intermediates, as these substances are precursors of the amino acids. This leads to the assumption that these mentioned parts of primary metabolism are stable throughout the different ecotypes. Secondary metabolites show strong negative correlations with enzymes of sugar metabolism among the considered Arabidopsis thaliana accessions. Variations of the levels of plant secondary metabolites are conceivable for accessions with different origin.