Abstract Background The interplay between metabolic processes and signalling pathways remains poorly understood. Global, detailed and comprehensive reconstructions of human metabolism and signalling pathways exist in the form of molecular maps, but they have never been integrated together. We aim at filling in this gap by integrating of both signalling and metabolic pathways allowing a visual exploration of multi-level omics data and study of cross-regulatory circuits between these processes in health and in disease. Results We combined two comprehensive manually curated network maps. Atlas of Cancer Signalling Network (ACSN), containing mechanisms frequently implicated in cancer; and ReconMap 2.0, a comprehensive reconstruction of human metabolic network. We linked ACSN and ReconMap 2.0 maps via common players and represented the two maps as interconnected layers using the NaviCell platform for maps exploration ([35]https://navicell.curie.fr/pages/maps_ReconMap%202.html). In addition, proteins catalysing metabolic reactions in ReconMap 2.0 were not previously visually represented on the map canvas. This precluded visualisation of omics data in the context of ReconMap 2.0. We suggested a solution for displaying protein nodes on the ReconMap 2.0 map in the vicinity of the corresponding reaction or process nodes. This permits multi-omics data visualisation in the context of both map layers. Exploration and shuttling between the two map layers is possible using Google Maps-like features of NaviCell. The integrated networks ACSN-ReconMap 2.0 are accessible online and allows data visualisation through various modes such as markers, heat maps, bar-plots, glyphs and map staining. The integrated networks were applied for comparison of immunoreactive and proliferative ovarian cancer subtypes using transcriptomic, copy number and mutation multi-omics data. A certain number of metabolic and signalling processes specifically deregulated in each of the ovarian cancer sub-types were identified. Conclusions As knowledge evolves and new omics data becomes more heterogeneous, gathering together existing domains of biology under common platforms is essential. We believe that an integrated ACSN-ReconMap 2.0 networks will help in understanding various disease mechanisms and discovery of new interactions at the intersection of cell signalling and metabolism. In addition, the successful integration of metabolic and signalling networks allows broader systems biology approach application for data interpretation and retrieval of intervention points to tackle simultaneously the key players coordinating signalling and metabolism in human diseases. Electronic supplementary material The online version of this article (10.1186/s12859-019-2682-z) contains supplementary material, which is available to authorized users. Keywords: Signalling, Metabolism, Networks, Comprehensive map, Systems biology, Cancer, Multi-level omics data, Data visualisation Background There is still a gap in understanding the coordination between metabolic functions and signalling pathways in mammalian cells. Metabolic processes and cell signalling pathways contain a large number of molecular species together with their complex relationships. No single mind can accurately account for all these molecular interactions whilst drawing conclusions from a process of descriptive thought. To tackle the complexity of these multi-molecular interactions networks, a systems biology approach is needed. In addition, there is a high number of omics data such as transcriptome, proteome, metabolome, etc. accumulated for many human diseases as age-related disorders (e.g. neurodegeneration or cancer). Modelling and interpretation of these data combining metabolic and signalling networks together can help to decipher the mechanisms responsible for deregulations in human disorders by considering a broader range of molecular processes types. Much of the produced high-throughput molecular data in many medical and biological applications remain under-explored due to the lack of insightful methods for data representation in the context of formally represented biological knowledge. Carefully designed maps of complex molecular mechanisms such as the whole-cell reconstructions of human metabolism in ReconMap 2.0 [[36]1, [37]2] or the global reconstruction of cell signalling of cancer in ACSN [[38]3] potentially provide ways to better exploit existing and new multi-omics data, by overlaying it on top of large molecular maps. ACSN is a resource and a web-based environment that contains a collection of interconnected signalling network maps ([39]https://acsn.curie.fr). Cell signalling mechanisms are depicted on the maps at the level of biochemical interactions, forming a large network of 4600 reactions covering 1821 proteins and 564 genes and connecting several major cellular processes [[40]3]. ACSN is composed of 5 interconnected maps of major biological processes implicated in cancer. The maps are further divided into functional modules that represent signalling pathways collectively responsible for the execution of a particular process. In total, there are 52 functional modules in the ACSN resource (See Table [41]1 for terms definition). Each of these modules can be visualised in the context of the global ACSN map or accessed as individual maps. The Atlas is a “geographic-like” interactive “world map” of molecular interactions. ACSN is supported by NaviCell platform for easy map navigation and its annotations using Google maps™ engine. The logic of navigation as scrolling and zooming; features as markers, pop-up bubbles and zoom bar are adapted from the Google map. Finally, NaviCell includes a powerful module for data visualisation. Users can map and visualise different types of “omics” data on the NaviCell maps [[42]4, [43]5]. Table 1. Term definitions used in the paper Graphical standards and exchange formats  XML format Markup language that defines a set of rules for encoding documents used to store and transport data by describing the content in terms of what data is being described.  SBGN Systems Biology Graphical Notation (SBGN) is a standard graphical syntax for representation of biological processes and interactions. SBGN is compatible with multiple pathway drawing and analytical tools, [44]http://sbgn.github.io/sbgn/  SBML Systems Biology Markup Language (SBML) is a representation format, based on XML, for communicating and storing computational models of biological processes. It is a free and open standard language with widespread software support, [45]http://sbml.org  Standard identifier (ID) Community-accepted nomenclature for scientific naming of biomolecules as genes, proteins, chemicals, drugs etc. The sources for standard IDs are repositories as UNIPROT, CHEB, HUGO, [46]http://identifiers.org  Data and models exchange formats Standard formats for data and models to facilitate networks and software intercompatibility. There are two major standard networks exchange formats, BIOPAX for complex networks and SIF for simple binary interactions. The CellDesigner xml format is a commonly-used exchange format compatible with multiple network analysis tools. Signalling and metabolic network map  Map Diagram of detailed molecular interactions with meaningful layout reflecting a certain biological process, which is graphically represented in CellDesigner tool.  Map module (in ACSN) Part of the map representing a sequence of molecular interactions responsible for execution of a particular function.  Metabolic pathway Subsystem (in ReconMap 2.0) Set of reactions forming a metabolic function. Set of metabolic reactions associated to (representing) a specific metabolic pathway.  Map node Graphical representation of a molecule on the map.  Map entity and alias Unique representation of a molecule on the map. As each molecule can be present multiple times on a map, each individual representation is called an alias of the entity. This definition corresponds to the CellDesigner feature. Networks merging procedure  Voronoi cell Individual shape allocated to a seed from the Voronoi method. Each cell is a space that contains only the seed and can be used to generate points inside without overlapping with close seeds.  Voronoi tessellation A partitioning of a plane into regions based on distance to points in a specific subset of the plane. That set of points (called seeds) is specified beforehand, and for each seed there is a corresponding region consisting of all points closer to that seed than to any other. These regions are called Voronoi cells. In our case, each seed is a molecule or a reaction’s central glyph.  Centroid Barycenter of a cluster.  Merging function of BiNoM Function allowing taking two or more CellDesigner maps and merging them in one unique map. This function modifies each entities’ id and alias but keeps the name, coordinates and notes. NaviCell  Semantic zoom A mechanism providing several map views with different levels of details depiction achieved by gradual exclusion of details while zooming out. It simplifies navigation through large maps of molecular interactions by providing several levels of details, resembling navigation through geographical maps. Exploring the map from a detailed toward a top-level view is achieved by gradual exclusion and modification (simplification and abstraction) of details. One of the main principles of semantic zooming is in that every detail which is shown on the map at a current zoom level, should be readable.  Marker Symbol indicating location of chosen objects on the map; adapted from Google maps.  Pop-up bubble Small window that opens by clicking on marker. Contains short description and hyperlinks related to the marked entity.  Annotation post Detailed map entity annotation created in CellDesigner by map manager. The annotation is converted to Annotation post and displayed in the associated blog by NaviCell. [47]Open in a new tab The manually curated genome-scale reconstruction Recon2.04 is a representation of the human metabolism. It accounts for 1733 enzyme-encoding genes associated to 7440 reactions which are distributed in 100 subsystems, referring to metabolic pathways. Furthermore, Recon2.04 accounts with 2626 unique metabolites distributed over eight cellular compartments [[48]2]. Subsequently, to visualise the resource, a comprehensive metabolic map termed ReconMap 2.0 was generated from the Recon2.04 resource [[49]1]. In the ReconMap 2.0 reactions (hyper-edges) were manually laid out using the biochemical network editor CellDesigner [[50]6]. ReconMap 2.0 is currently distributed in a Systems Biology Graphical Notation (SBGN) compliant format and its content is also accessible via a web interface ([51]https://vmh.uni.lu/#reconmap). All major human metabolic pathways are considered and represented as a seamless network where different pathways are interconnected via common molecules. There are 96 subsystems on the ReconMap 2.0, each of them representing a specific metabolic pathway (See Table [52]1 for terms definition). By integrating these resources together, it will be possible to elucidate the crosstalk between metabolic and signalling networks. In addition, the integrated networks, provided in a common graphical language and available in standard exchange formats, makes them accessible for multiple systems biology tools. It opens an opportunity to model coordination between signalling pathways and metabolism using various systems biology approaches. Among others, there are several methods for multi-level omics data analysis in the context of the biological network maps that allow defining “hot” areas in molecular mechanisms and point to key regulators in physiological or in pathological situations [[53]7–[54]9] and beyond. General workflow for integration of ACSN and ReconMap 2.0 networks With the aim to integrate signalling and metabolic networks there is a need to find common players (proteins) that participate in the regulation of metabolic processes and simultaneously involved in signal transduction pathways. Thus, the networks can be interconnected via these common players. In addition, some solution for visualisation of proteins participating in the catalytic process in ReconMap 2.0 should be provided, since there is no such representation up to date. The rationale behind the proposed methodology is to take an advantage of the CellDesigner SBML format for networks representation and develop a robust automated algorithm for an efficient finding of coordinates for new entities avoiding an overlap with existing elements and visualising these entities in the vicinity of the corresponding reactions they regulate. The integrated networks can be provided as interconnected layers supported by NaviCell platform for navigation and data integration. The suggested methodology is applied for ACSN and ReconMap 2.0 resources integration. However, this is a generic method applicable for integration of different types of networks prepared in CellDesigner SBML format (Fig. [55]1). In the following sections of the paper, we explain the challenges and describe how each step mentioned in the workflow was addressed. Fig. 1. [56]Fig. 1 [57]Open in a new tab General workflow for integration of proteins into a metabolic network. (1) Extraction of the informations on proteins present in metabolic reactions from a model and CellDesigner file. (2) Addition of proteins in the vicinity of catalysed reactions. (3) Merging of obtained proteins with the metabolic map through the BiNoM plugin. (4) As a result, a CellDesigner network file containing proteins on top of the original metabolic network is obtained. This file can be later integrated into NaviCell through the NaviCell Factory tool The workflow in the Section 2 includes the following major steps (see Table [58]1 for terms definition): * Identification of common proteins between ACSN and ReconMap 2.0 networks * Finding metabolic and molecular processes crosstalk between ACSN and ReconMap 2.0 * Displaying protein nodes on the ReconMap 2.0 map * ACSN-ReconMap 2.0 networks integration and visualisation using NaviCell Materials and methods Step-by-step procedure for network integration Identification of common proteins between ACSN and ReconMap 2.0 networks ACSN and ReconMap 2.0 maps contain information on proteins implicated in the regulation of reactions. First, the systematic use of the common identifiers as standard protein names (HUGO) for all proteins in both resources was verified and inconsistencies corrected. Thus, the proteins found in both resources ACSN and ReconMap 2.0, were compared, quantified and visualised. We detected 252 proteins in common between the two networks (Additional file [59]1). Displaying protein nodes on the ReconMap 2.0 map ACSN and ReconMap 2.0 are both used as visual objects for exploration of processes as well as for data integration and visualisation in the context of the maps. After identification of the cross-talks between the two resources, it is important to ensure that all components of the maps are represented in a visual manner suitable for meaningful visualisation of omics data. Due to the different nature of the networks, protein nodes are explicitly visualised on the ACSN map. However, in the ReconMap 2.0 the Standard Names (Identifiers) of proteins regulating metabolic reactions are included into the reaction annotations, but not represented visually on the map canvas. This precludes visualisation of omics data in the context of ReconMap 2.0 map. We developed a procedure for displaying the protein nodes on the ReconMap 2.0 map in the vicinity of the corresponding reaction edges, that now permits a multi-omics data visualisation in the context of both ACSN and ReconMap 2.0 layers. Extraction of information regarding reactions and implicated genes in the metabolic network * Recuperation of the information from the Recon2.04 model + ReconMap 2.0 is the graphical representation of the literature-based genome-scale metabolic reconstruction Recon2.04, which is freely available at ([60]https://vmh.uni.lu/#downloadview). It is stored as a MatLab “.mat” file which contains a direct link between metabolic reactions and gene Entrez, specified by gene-rules. Therefore, it is possible to generate a direct protein-reaction association based on the gene codifying for the protein. As ACSN uses HUGO Standard Identifiers, Entrez IDs in ReconMap 2.0 were first converted to HUGO. + It is important to stress that this approach is based on a simplified assumption that if a protein is associated to a metabolic reaction in ReconMap 2.0, it may have a role in catalysis of the reactions. However, it is clear that the biological regulation is much more sophisticated than this basic assumption. For example, there are many protein complexes collectively regulating propagation of metabolic reaction and only part of them are actual enzymes that execute the catalysis, whereas others are co-factors of regulatory sub-units. Moreover, the activation states of proteins that is often regulated by post-translational modifications are also not taken into account in this simplified approach. * Recuperation of entities positions in ReconMap 2.0 from the XML network file + In the graphical representation of reactions in CellDesigner, each reaction contains a central glyph in the form of a square. This glyph is normally used to allocate the position of the markers (see Table [61]1 for terms definition). However, its location is not explicitly saved in the network XML file. A specific function of NaviCell factory can calculate the coordinates of these glyphs and extract them in a separated file. These coordinates can be later used as a reference positions to assign protein nodes position in the ReconMap 2.0 map canvas. Automated calculation of proteins coordinates in vicinity of corresponding reactions at ReconMap 2.0 network * Computing Voronoi cells for all elements + By using the Voronoi method, each element of the network (molecules, reaction glyphs, etc.) is associated to a Voronoi cell. This method guarantees the lack of overlapping elements with already existing entities in the network when adding new proteins (Fig. [62]2). * Creation of randomly distributed points inside each reaction’s Voronoi cell + When each entity has a cell assigned, cells of reactions’ central glyphs are utilised. Each cell has a certain number of points assigned randomly inside the cell. For our purpose, 100 points were deemed sufficient (Fig. [63]2). * Application of K-means algorithm to create K clusters + Each reaction has a certain number of proteins implicated in its catalysis. Using the information from the model, the K-means algorithm was applied to identify the number of cluster centres corresponding to the number of protein nodes (Fig. [64]2). * Assigning protein positions using centroids coordinates of each cluster + After the protein clusters are found, their centroids (see Table [65]1 for terms definition) are calculated and saved as the coordinated of the proteins tied to the specific reaction as catalysts (Fig. [66]2). Fig. 2. [67]Fig. 2 [68]Open in a new tab Illustration of the three steps for automated proteins addition in the vicinity of a reaction. The first step is to generate a Voronoi cell for each entity in the map. The second step is to generate several randomly assigned points in the Voronoi cell of reactions catalysed by proteins. The third step consists in using the k-means algorithm to generate the needed number of clusters and assign the cluster’s centroids coordinates as those of the proteins catalysing the reaction in question Conversion of obtained coordinates into a standard format (SBML) * Saving protein positions in a BiNoM Reaction Format + Following the previous steps, a file in the BiNoM Reaction Format is obtained, containing the name of the proteins as well as their coordinates and sizes. This simple file will then be converted to a standard CellDesigner SBML format to be compatible with the original metabolic network. As CellDesigner allows the manipulation of “aliases” (multiple copies of the same entity); each protein with the same name present multiple times will have an apostrophe attached to its name based on the number of its repetition within the network. * Conversion of BiNoM Reaction Format into a CellDesigner map + Using a custom python script, information stored in the BiNoM Reaction Format is transformed into a XML file following the SBML format. This file will contain each protein names, IDs, alias IDs, coordinates and type. As for now, only the manipulation of simple proteins is available. * Merging of the ReconMap 2.0 and Proteins maps using BiNoM merging function. + Once the file containing proteins to add to the metabolic map is obtained, as they both are in the same SBML format, it is possible to merge them by using a function of the BiNoM plugin. This function allows transforming two or more separated maps into one unique map. This final merged map will be transformed into the NaviCell environment using the NaviCell Factory package ([69]https://github.com/sysbio-curie/NaviCell). Thus, proteins implicated in the catalysis of a reaction can be seen in the vicinity of the corresponding reactions (Additional file [70]2, Fig. A). It is important to note that in some cases, reactions are regulated by many proteins, for example in the case of protein families, and the resulting configuration of protein nodes can be very dense (Additional file [71]2, Fig. B). This aspect can be improved by grouping protein families and visualising them together as a single generic entity. However, it is not always relevant to group all protein sharing a similar name by “family”, since different family members might fulfil distinct or even opposite function, leading to a misinterpretation of the omics data in the context of the maps. Therefore, each protein was kept as a unique and independent entity. Thanks to this method, 1.550 proteins were allocated in the ReconMap 2.0 canvas associated to more than 7.500 aliases. The algorithm for assigning proteins’ coordinates is robust and its computation time is also scalable as the generation of the 7.500 allocation points is resolved in a matter of seconds. ACSN and ReconMap 2.0 merging Once the protein positions file has been generated, it was converted to a CellDesigner [[72]10, [73]11] XML format through a custom python script ([74]https://github.com/sysbio-curie/CellDesigner_networks_map_integrat ion_procedure). This script allows to obtain a file in XML format following the standard of CellDesigner’s SBML. This ‘map’ contains only proteins in the positions they should belong on the final metabolic map. This file was then merged with the ReconMap 2.0 network by using an existing merging function of BiNoM [[75]12, [76]13] to obtain the final network containing the original ReconMap 2.0 as well as the proteins in the vicinity of reactions they catalyse. Tools, data source and code accessibility Maps generation tool CellDesigner [[77]10, [78]11] is a tool used for the construction of both networks and its standard notation allowed the integration and linking across these maps. Both maps are available in a XML format, thus facilitating their automated manipulation. Map entity annotation with NaviCell format The annotation panel followed the NaviCell annotation format of each entity and reaction of the maps includes sections ‘Identifiers’, ‘Maps_Modules’, ‘References’ and ‘Confidence’ as detailed in [[79]3].