Application of Atlas of Cancer Signalling Network in preclinical studies

Application of Atlas of Cancer Signalling Network in preclinical studies Abstract Cancer initiation and progression are associated with multiple molecular mechanisms. The knowledge of these mechanisms is expanding and should be converted into guidelines for tackling the disease. Here, we discuss the formalization of biological knowledge into a comprehensive resource: the Atlas of Cancer Signalling Network (ACSN) and the Google Maps-based tool NaviCell, which supports map navigation. The application of ACSN for omics data visualization, in the context of signalling maps, is possible via the NaviCell Web Service module and through the NaviCom tool. It allows generation of network-based molecular portraits of cancer using multilevel omics data. We review how these resources and tools are applied for cancer preclinical studies. Structural analysis of the maps together with omics data helps to rationalize the synergistic effects of drugs and allows design of complex disease stage-specific druggable interventions. The use of ACSN modules and maps as signatures of biological functions can help in cancer data analysis and interpretation. In addition, they empowered finding of associations between perturbations in particular molecular mechanisms and the risk to develop a specific type of cancer. These approaches are helpful, among others, to study the interplay between molecular mechanisms of cancer. It opens an opportunity to decipher how gene interactions govern the hallmarks of cancer in specific contexts. We discuss a perspective to develop a flexible methodology and a pipeline to enable systematic omics data analysis in the context of signalling network maps, for stratifying patients and suggesting interventions points and drug repositioning in cancer and other diseases. systems biology, cancer, signalling network maps, omics data integration and analysis, data visualization, synthetic interactions, drug response Introduction According to the current scientific understanding, different signalling pathways interact to create a complex network featuring feed forward loops, backward regulatory loops and many alternative paths that support redundancy. It is generally believed that, under most pathological transformations, cells do not exploit new molecular signalling mechanisms but rather hijack existing molecular programs. This affects not only intracellular functions but also the interactions between different cell types, leading to a new, yet pathological, status of the system. It is likely that a specific combination of molecular characteristics dictates specific cell signalling states, maintaining the pathological disease status. Identifying and manipulating the key molecular players that control these cell signalling states, and shifting the pathological cell state towards the desired healthy or, alternatively, lethal phenotype, are major challenges for molecular biology both in the more general context of human diseases and in the specific context of cancer [1–3]. To allow an appropriate data analysis of the molecular mechanisms supporting carcinogenesis, the information on these mechanisms should be systematically and adequately represented. The knowledge concerning the molecular signalling mechanisms of the cell is spread across thousands of publications, mostly in a human-readable, and computer-unfriendly, form, precluding the direct application of many bioinformatics and systems biology approaches. Hence, it is important to compile such knowledge in a computer-readable form. The most common way to approach this problem is to represent the relationships between molecules in a network form, resulting in pathway diagrams, which can be found in various pathway databases [4]. As the amount of information about biological mechanisms steadily increases, introducing a different approach to organize and structure this information is essential. Our aim is to provide a global cell signalling picture with sufficient granularity to preserve molecular details, capturing crosstalks and feedback loops between molecular circuits. For this purpose, comprehensive signalling network maps covering simultaneously multiple cellular processes are more suitable than disconnected pathway diagrams. This review focuses on different approaches for a rational representation of cell signalling in cancer, and will use the Atlas of Cancer Signalling Network (ACSN) as an example. Visualization and analysis of omics data in the context of signalling networks facilitate data interpretation and can help in highlighting deregulated mechanisms. Furthermore, data analysis in the context of signalling networks can help detecting patterns in the data projected onto the molecular mechanisms represented in the signalling maps, providing information on enriched functional modules (‘hot’ deregulated areas), key players and ‘bottleneck’ points [3, 5]. Correlating the status of those network variables with the phenotype, as drug resistance or patient survival, can be followed by clustering approaches to allow us stratifying patients according to their network-based molecular portraits. Moreover, these approaches can help in suggesting intervention points and in designing appropriate therapeutic schemes [6]. As a productive idea for intervention schemes, synthetic lethality (SL) provides a conceptual framework to develop cancer-specific drugs. This classical paradigm defines SL interactions as a phenomenon where combinations of two gene deletions significantly compromise cell viability, whereas the deletion of one of those genes does not [7]. The idea of the SL treatment approach is to take advantage of the vulnerabilities in tumour cells, which can be characterized by the abnormal function of one of the genes from the SL pair. Targeting the SL partner allows then to selectively kill the tumour cells, avoiding or reducing side effects on normal cells [8]. The approach also provides the clinician with a biomarker to select patients that could respond to the treatment. Giving the complexity of signalling mechanisms simultaneously involved in cancer, the SL pair paradigm should be extended to the SL sets or combinations paradigm [9, 10]. The computational approaches that allow in silico testing of multiple synthetic interactions combinations, considering large comprehensive signalling networks and cancer omics data, are discussed in this review. ACSN: geographical map of molecular mechanisms Deregulation of molecular mechanisms leading to cancer can be observed in various processes such as cell cycle, cell death, DNA repair and replication, cell motility and adhesion, cell survival mechanisms, immune processes, angiogenesis, tumour microenvironment and others. Most of them are collectively or sequentially involved in tumour formation and modified as the tumour evolves. The scientific literature often suggests that in pathological situations, the normal cell signalling network is altered by deregulated coordination between pathways or disruption of existing molecular pathways, rather than by creating completely new signalling pathways. The most common abnormalities in pathological situations are perturbations at the gene expression level, protein abundance or protein post-translational modifications, irregular ‘firing’ or silencing of particular signals, wrong subcellular localization of particular molecules and so on. Such quantitative rather than qualitative network changes, compared with the normal cell signalling, could be studied in the context of comprehensive signalling networks by analysing experimental data obtained from tumour samples, patient-derived xenografts, cancer-related cell lines or animal models. This approach helps to understand the interplay between molecular mechanisms in cancer and to decipher how gene and protein interactions govern the hallmarks of cancer [11], in specific settings. Despite the existence of a large variety of pathway databases and resources [4], only few of them are cancer-specific, and none of these resources depict the processes with sufficient granularity. In addition, pathway browsing interfaces are becoming more important for cancer researchers and clinicians but require further improvements. Therefore, we constructed a resource, the ACSN (http://acsn.curie.fr) that aims at formalizing the knowledge on cancer-related processes in the form of a comprehensive signalling network map, for data interpretation in basic research and preclinical studies [12]. The construction and update of ACSN involve manual mining of the cellular and molecular biology literature, along with the participation of experts in those fields. ACSN differs from other databases because it contains a deep comprehensive description of cancer-related mechanisms, retrieved from the most recent literature, based on the hallmarks of cancer (Figure 1). Cell signalling mechanisms are depicted using the CellDesigner tool [13] at the level of biochemical interactions, assembling a large network of 4826 reactions covering 2371 proteins and based on approximately 3000 references (Table 1). Table 1. Content of ACSN (adapted from [12]) Feature Content Maps of biological processes 5 Functional modules 52 Chemical species 5975 Reactions 4826 Proteins 2371 Metabolites 595 Genes 159 References 2919 Feature Content Maps of biological processes 5 Functional modules 52 Chemical species 5975 Reactions 4826 Proteins 2371 Metabolites 595 Genes 159 References 2919 Table 1. Content of ACSN (adapted from [12]) Feature Content Maps of biological processes 5 Functional modules 52 Chemical species 5975 Reactions 4826 Proteins 2371 Metabolites 595 Genes 159 References 2919 Feature Content Maps of biological processes 5 Functional modules 52 Chemical species 5975 Reactions 4826 Proteins 2371 Metabolites 595 Genes 159 References 2919 Figure 1. View largeDownload slide Structure of ACSN resource. The scheme demonstrates the concept of ACSN construction starting from the cancer hallmarks: collecting information about molecular mechanisms underlying those hallmarks from scientific publications and manually depicting them in the global map of ACSN and further supporting by consulting the information from the external pathway databases. ACSN is hierarchically organized into three levels: the seamless global map divided into the interconnected biological process maps that are further decomposed into interconnected module maps. ACSN can be exploited through Web-based NaviCell interface allowing map navigation using Google Maps engine, map commenting via associated blog system and user omics data visualization and analysis (Adapted from [12]). Figure 1. View largeDownload slide Structure of ACSN resource. The scheme demonstrates the concept of ACSN construction starting from the cancer hallmarks: collecting information about molecular mechanisms underlying those hallmarks from scientific publications and manually depicting them in the global map of ACSN and further supporting by consulting the information from the external pathway databases. ACSN is hierarchically organized into three levels: the seamless global map divided into the interconnected biological process maps that are further decomposed into interconnected module maps. ACSN can be exploited through Web-based NaviCell interface allowing map navigation using Google Maps engine, map commenting via associated blog system and user omics data visualization and analysis (Adapted from [12]). Currently, ACSN contains representations of molecular mechanisms that are frequently dysregulated in cancer, such as cell cycle, DNA repair, cell death, cell survival and epithelial to mesenchymal transition (EMT). Cell signalling mechanisms are depicted in the maps in great detail, creating together a map of molecular interactions, presented as a global ‘geographic-like’ representation (Figure 1A). ACSN has a hierarchical structure, composed of interconnected maps of processes altered in cancer. Each map is further divided into functional modules, corresponding mainly to canonical signalling pathways (Figures 1 and 2C). Figure 2. View largeDownload slide Browsing interface of ACSN. (A) ACSN interface with selection panel and data visualization menu. Querying ACSN is possible via the search window or by checking on the entity in the list of entities. Distribution of frequently mutated oncogene Myc proto-oncogene protein (MYC) across molecular mechanisms on the ACSN maps is indicated; (B) Google Maps-like features of NaviCell for visualization and annotation of map entities (markers, call-out with links to external databases, citations and ACSN maps of and functional modules where MYC protein is found); (C) Zoom in on Wnt non-canonical module of cell survival map to observe signalling processes where MYC protein is involved (Adapted from[12]). Figure 2. View largeDownload slide Browsing interface of ACSN. (A) ACSN interface with selection panel and data visualization menu. Querying ACSN is possible via the search window or by checking on the entity in the list of entities. Distribution of frequently mutated oncogene Myc proto-oncogene protein (MYC) across molecular mechanisms on the ACSN maps is indicated; (B) Google Maps-like features of NaviCell for visualization and annotation of map entities (markers, call-out with links to external databases, citations and ACSN maps of and functional modules where MYC protein is found); (C) Zoom in on Wnt non-canonical module of cell survival map to observe signalling processes where MYC protein is involved (Adapted from[12]). The navigation interface includes features such as scrolling, zooming, markers and callouts using the Google Maps technology adapted by NaviCell [14] (Figure 3), Web-based platform supporting ACSN and similar efforts in CellDesigner format [15, 16] or other formats [17]. The semantic zooming in NaviCell (http://navicell.curie.fr) provides several view levels, achieved by gradual exclusion of details and abstraction of information on zooming out (Figure 2B). Figure 3. View largeDownload slide General architecture of NaviCell environment (Adapted from [18]). Figure 3. View largeDownload slide General architecture of NaviCell environment (Adapted from [18]). ACSN is a unique resource of cancer signalling knowledge, with an enormous amount of information embedded and organized. Together with NaviCell, it is optimized for integration and visualization of cancer molecular profiles generated by high-throughput technologies, drug screening data or synthetic interactions studies. The integration and analysis of these data in the context of ACSN may help to better understand the biological relevance of results, guiding scientific hypotheses and suggesting potential therapeutic intervention for cancer patients. In addition, as ACSN covers major cell signalling processes, the resource and associated methods for data analysis using ACSN are suitable for applications in many biological fields and for studying various human diseases. The Atlas is currently being extended with additional maps depicting molecular mechanisms of DNA replication, telomere maintenance, angiogenesis, immune response and others that will be integrated in future releases. The Atlas will not only cover intracellular processes but also crosstalks of cancer cell with the components of tumour microenvironment. An additional level of complexity will be added to the Atlas in the near future, representing different types of cells surrounding the tumour, and their interplay, to enable modelling of complex phenotypes. Molecular portraits of cancer: data visualization and analysis using ACSN Data visualization in NaviCell Web Service environment The data integration into ACSN is possible through NaviCell Web Service, a user-friendly interface making a part of the NaviCell tool [18]. It allows uploading several types of ‘omics’ data—e.g. mRNA expression data, microRNA, proteins, mutation profiles and copy number data—in simple text table (tab-separated) format. The data can be visualized simultaneously in the context of molecular interaction maps (Figure 4). Detailed instructions, tutorials and life examples of multi-omics data visualization using NaviCell Web Service are available at https://navicell.curie.fr/pages/nav_web_service.html. Figure 4. View largeDownload slide General architecture of NaviCell Web service server. Client software (light blue layer) communicates with the server (red layer) through standard HTTP requests using the standard JSON format to encode data (RESTful Web service, dark blue layer). A session (with a unique ID) is established between the server and the browser (yellow layer) through Ajax communication channel to visualize the results of the commands send by the software client (Adapted from [18]). Figure 4. View largeDownload slide General architecture of NaviCell Web service server. Client software (light blue layer) communicates with the server (red layer) through standard HTTP requests using the standard JSON format to encode data (RESTful Web service, dark blue layer). A session (with a unique ID) is established between the server and the browser (yellow layer) through Ajax communication channel to visualize the results of the commands send by the software client (Adapted from [18]). Depending on the type of data, different visualization modes can be applied to obtain an informative picture, such as bar plots, glyphs and map staining (Table 2). The data can be visualized at different zoom levels. Sample annotation files uploaded with the data can serve to define groups of samples. Table 2. Data display modes in NaviCell and NaviCom (adapted from [19]) Data type Visualization on mode Data display Units mRNA expression Map staining Level Gene copy number Heat map Count Muta on data Glyph 1 Frequency Methyla on data Glyph 2 Intensity miRNA expression Glyph 3 Level Protein expression Glyph 4 Level Data type Visualization on mode Data display Units mRNA expression Map staining Level Gene copy number Heat map Count Muta on data Glyph 1 Frequency Methyla on data Glyph 2 Intensity miRNA expression Glyph 3 Level Protein expression Glyph 4 Level Table 2. Data display modes in NaviCell and NaviCom (adapted from [19]) Data type Visualization on mode Data display Units mRNA expression Map staining Level Gene copy number Heat map Count Muta on data Glyph 1 Frequency Methyla on data Glyph 2 Intensity miRNA expression Glyph 3 Level Protein expression Glyph 4 Level Data type Visualization on mode Data display Units mRNA expression Map staining Level Gene copy number Heat map Count Muta on data Glyph 1 Frequency Methyla on data Glyph 2 Intensity miRNA expression Glyph 3 Level Protein expression Glyph 4 Level A novel mode of data visualization for continuous data (e.g. expression) provided by the NaviCell Web Service is the ‘map staining’. With this technique, the values mapped to individual molecular entities or group of entities (e.g. score of functional module activities, see below) result in a colourful background of the network map [18]. All the approaches for data integration into the signalling maps described above allow to elucidate and interpret the omics data, compare samples or groups, find patterns across the molecular mechanisms depicted on the maps, grasp deregulated ‘hot areas’ on the maps and major involved players and draw hypotheses as to which mechanisms to focus on. These signalling network-based molecular signatures thus help to stratify patients (Figure 5). Figure 5. View largeDownload slide Breast cancer gene expression data integration and analysis using NaviCell. The mRNA expression data from TCGA collection has been used for evaluation of functional modules activities and ACSN colouring as ‘map staining’ for (A) basal-like and (B) luminal A breast cancer types. The two breast cancer subtypes are characterized by different patterns of module activities. (Adapted from [12]). Figure 5. View largeDownload slide Breast cancer gene expression data integration and analysis using NaviCell. The mRNA expression data from TCGA collection has been used for evaluation of functional modules activities and ACSN colouring as ‘map staining’ for (A) basal-like and (B) luminal A breast cancer types. The two breast cancer subtypes are characterized by different patterns of module activities. (Adapted from [12]). Various omics data are available in the public and commercial repositories [20]. However, there is a lack of tools supporting the integration of big data sets from these databases and the visualization on signalling network maps in an efficient way and with optimized visualization settings. To answer to this demand, NaviCom has been developed, a python package providing a Web interface for simultaneous display of multilevel data in the context of signalling network maps [19]. NaviCom (http://navicom.curie.fr) provides a bridge between the cBioPortal database and the NaviCell interactive tool for data visualization. NaviCom integrates functionalities from the cBioFetchR R package to import high-throughput data sets from cBioPortal to NaviCell and the Navicom Python module, allowing automatized simultaneous visualization of multilevel omics data on the interactive signalling network maps provided by the NaviCell environment (Figure 6A). NaviCom proposes several standardized modes of data visualization on signalling network maps to address specific biological questions (Table 2). Figure 6. View largeDownload slide General architecture of NaviCom environment. The NaviCom interface provides the user with an updated list of studies from cBioPortal and links to ACSN and NaviCell maps collections. When visualization is launched, NaviCom starts a new NaviCell session and calls a cgi on the server. The cgi downloads cBioPortal data to the NaviCell session and displays them to generate the molecular portrait selected by the user (Adapted from [19]). Figure 6. View largeDownload slide General architecture of NaviCom environment. The NaviCom interface provides the user with an updated list of studies from cBioPortal and links to ACSN and NaviCell maps collections. When visualization is launched, NaviCom starts a new NaviCell session and calls a cgi on the server. The cgi downloads cBioPortal data to the NaviCell session and displays them to generate the molecular portrait selected by the user (Adapted from [19]). This tool enables the generation of complex molecular portraits from multiple omics data sets from cBioPortal. Detailed instructions and tutorials are available at https://navicom.curie.fr/tutorial.ph. In the near future, the NaviCom platform will be extended and will provide access to many types of omics data from a wide range of databases (TCGA, ICGC, HGMB, METABRIC and CCLE). In addition, to allow for a wider description of the molecular mechanisms implicated in the studied sample, signalling networks available in databases as KEGG [21], Reactome [22] and others, will be also integrated and used for high-throughput data analysis via the NaviCom platform. Data analysis using ACSN To identify dysregulated signalling pathways of functional modules in a molecular map from a given data set, several tools are currently available. Gene Set Enrichment Analysis (GSEA) (http://software.broadinstitute.org/gsea/index.jsp) is a computational method aimed at finding overrepresented modules in a ranked gene list, using a weighted Kolmogorov–Smirnov test [23]. ACSNMineR (https://github.com/sysbio-curie/ACSNMineR) is an R package that incorporates ACSN information to calculate enriched or depleted modules by means of a Fisher exact test or a hypergeometric test [24]. The Representation and Quantification of Module Activity (ROMA) method, implemented in Java and R (https://github.com/sysbio-curie/Roma, https://github.com/sysbio-curie/rRoma), is designed for fast and robust computation of the activity of gene sets (or modules) with coordinated expression. ROMA uses the first component of principal component analysis to summarize the co-expression of a group of genes in a gene set. ROMA also proves additional functionalities: (i) calculation of the individual gene contribution to the module activity level and determination of the genes that are contributing the most to the first principal component, (ii) several variants of computation for the first principal component, i.e. weighted and centred methods and (iii) estimation of the statistical significance of the proportion of variance explained by the first principal component, as well as the spectral gap between the variance explained by the first and second component (representing the homogeneity for the gene set) [25]. The module activity scores calculated by these methods can be visualized in the context of ACSN using the map staining technique as described above (Figure 5). Such visualization is automated in ACSNMineR and ROMA. ACSN was also used as a source of module definitions for benchmarking the DeDaL tool. DeDaL allows the creation of a mixed data-driven and structure-driven network layout, which can be more insightful for grasping the correlation patterns in the multivariate data on top of the networks [26]. ACSN module definitions were applied for testing a method for inferring hidden causal relations between pathway members using reduced Google matrix of directed biological networks [27]. ACSN acts as a source of functional module definitions and protein–protein interactions network in data analysis projects, especially those from cancer data. For example, gene lists from functional modules of the DNA repair map were used to study homologous recombination (HR) deficiency in invasive breast carcinomas [28]. ACSN was used as a source of signatures for processes involved in cancer for classification of gene signatures and generation of InfoSigMap, an interactive online map showing the structure of compositional and functional redundancies between signatures from various sources [17]. Exploiting the ACSN in preclinical research Explaining the synergistic effect of combined treatments in breast cancer DNA repair inhibitors are holding promises to improve cancer therapy, but their application is limited by the compensatory activities of different repair pathways in cancer cells. For example, PARP inhibitors, which act as SL with BRCA deficiency, appear less efficient in patients with active HR repair mechanisms [29]. During treatment, some tumours escape the elimination through compensatory mutations that restore the HR activity or stimulate the activity of alternative repair pathways such as non-homologous end joining (NHEJ) and alternative non-homologous end joining (Alt-NHEJ). A new class of DNA repair pathways inhibitors (Dbait or AsiDNA, a Dbait derivative) has been recently developed, consisting of 32 bp deoxyribonucleotides DNA double helix that mimics double-strand breaks (DSB). It acts as an agonist of DNA damage signalling, thereby inhibiting DNA repair enzyme recruitment at the damage site [30]. However, studies on the effects of Dbait on multiple types of cancer cell lines show occurrences of resistance in a cancer type-independent manner. Depending on the genetic background, different breast cancer tumours vary in their sensitivity to DNA repair inhibitors, as PARP inhibitors and AsiDNA. To understand the molecular mechanisms underlining these differences, a combination of experimental and bioinformatics approaches was applied. Triple negative breast cancer (TNBC) cell lines were studied for their sensitivity to AsiDNA, the derivative of Dbait DNA repair inhibitor, and Olaparib, the PARP inhibitor. Different TNBC cell lines show a wide distribution of response/resistance to these drugs, despite the fact that these cell lines are related to the same disease type. Integrative analyses of omics data from these cell lines encompassing mRNA expression, copy number variations and mutational profiles were performed, retrieving non-overlapping unique gene sets robustly correlated with resistance to each one of the drugs. Analysis of the omics data in the context of ACSN maps highlighted dysregulated functional modules across ACSN, associated with resistance to each one of the drugs allowing to establish drug resistance network-based molecular portraits. This analysis confirmed that different specific defects in DNA repair machinery are associated to AsiDNA (Figure 7A) or Olaparib (Figure 7B) resistance. Importantly, it showed involvement of different compensatory DNA repair mechanisms in cell lines resistant to AsiDNA when compared with cell lines resistant to Olaparib (Figure 7D), suggesting a rationale to combine these two drugs. The authors confirmed a synergistic therapeutic effect of the combined treatment with AsiDNA and PARP inhibitors in TNBC, while sparing healthy tissue (Figure 7C) [31]. Figure 7. View largeDownload slide Overcoming resistance of TNBC cell lines to DNA repair inhibitors. Molecular portraits of TNBC cell lines resistant to (A) AsiDNA or (B) Olaparib, visualized on DNA repair map. (C) Cell survival to combination of AsiDNA and Olaparib, with AsiDNA 1(black line), without AsiDNA (grey line); dashed lines indicate calculated cell survival for additive effect of two drugs. (D) Schematic representation of inhibitory mechanisms of AsiDNA and Olaparib. Base excision repair (BER), HR, NHEJ, Alt-NHEJ (Adapted from [31]). Figure 7. View largeDownload slide Overcoming resistance of TNBC cell lines to DNA repair inhibitors. Molecular portraits of TNBC cell lines resistant to (A) AsiDNA or (B) Olaparib, visualized on DNA repair map. (C) Cell survival to combination of AsiDNA and Olaparib, with AsiDNA 1(black line), without AsiDNA (grey line); dashed lines indicate calculated cell survival for additive effect of two drugs. (D) Schematic representation of inhibitory mechanisms of AsiDNA and Olaparib. Base excision repair (BER), HR, NHEJ, Alt-NHEJ (Adapted from [31]). Complex stage-specific interventions in MAPK pathway to disrupt proliferative signalling in bladder cancer The idea of an SL treatment approach is to take advantage of the peculiarities of tumour cells having an abnormal expression or functionality of one gene from an SL pair. Targeting another SL partner allows selective killing of the tumour cells [32]. This approach is applied in BRCA2 mutated breast cancer cases using PARP inhibitors. However, there is a frequent escape from the treatment, requiring a more complex approach. Treatment failure can be because of the robustness of cell signalling network ensured by redundant mechanisms that provide the possibility to bypass the effect of drugs [33]. Therefore, the ways for identifying and blocking those active compensatory pathways should be found. One of the approaches to solve this issue is by taking into account the signalling network structure to find the most optimal SL gene combinations, possibly more than a pair [10, 34, 35]. A computational strategy to suggest complex intervention sets has been developed and demonstrated using the mitogen-activated protein kinase (MAPK) signalling network [36]. The MAPK signalling network is coordinated with various processes implicated in cell survival, and currently included into the Cell Survival map of ACSN. The strategy involves two steps: (i) identification of tumour stage-specific active functional modules, i.e. sets of MAPK signalling network components that are transcriptionally deregulated in bladder cancer [37] compared with normal samples, and (ii) computation of intervention sets of MAPK map components, whose disruption block all the proliferative paths fostered by the identified active functional modules in bladder cancer [38] (Figure 8A). Figure 8. View largeDownload slide Computational strategy for finding stage-specific interventions sets using detailed reaction network analysis and omics data. (A) Detailed MAPK network map is shown with schematically indicated activated modules (see text). Finding minimal hitting sets allows to cut all paths (schematically shown by arrows) from the activated modules to the proliferation phenotype. (B) Stage-specific activated modules detected in MAPK network using bladder cancer transcriptome data. Module enrichment score were computed by GSEA method. The most contributing leading-edge genes with highest differential expression level are highlighted by red. The optimal hitting set lists from these gene elements of MAPK network, which on removal cuts all the paths from the corresponding activated modules to the proliferation phenotype, were calculated using the OCSANA algorithm, and intervention sets for each stage of bladder cancer were suggested (see text) (Adapted from [39]). Figure 8. View largeDownload slide Computational strategy for finding stage-specific interventions sets using detailed reaction network analysis and omics data. (A) Detailed MAPK network map is shown with schematically indicated activated modules (see text). Finding minimal hitting sets allows to cut all paths (schematically shown by arrows) from the activated modules to the proliferation phenotype. (B) Stage-specific activated modules detected in MAPK network using bladder cancer transcriptome data. Module enrichment score were computed by GSEA method. The most contributing leading-edge genes with highest differential expression level are highlighted by red. The optimal hitting set lists from these gene elements of MAPK network, which on removal cuts all the paths from the corresponding activated modules to the proliferation phenotype, were calculated using the OCSANA algorithm, and intervention sets for each stage of bladder cancer were suggested (see text) (Adapted from [39]). The procedure was applied to five different bladder cancer stages [37]. Differential gene expression levels were computed, relative to healthy conditions, using transcriptomic data. Regions of the map having high density of differentially expressed genes/proteins were identified and scored by GSEA [23]. The highly scoring regions were assumed to likely point to sources of proliferative signals in the tumour. If paths exist in the map from the components belonging to any strongly activated network region to the node ‘Proliferation’, then presumably the region contributes to the activation of cell division (Figure 8A). The removal of a set of proteins from the network might block all such proliferative paths. This idea was formalized by the notion of the minimal cut sets, which were computed using the OCSANA algorithm in the Cytoscape plugin BiNoM [40]. The OCSANA algorithm computes the minimal cut sets, by simultaneously prioritizing them with respect to the potential effect on the target network nodes, while avoiding side effects on the parts of the network that functionally should be preserved. In the less invasive Ta stage, two significantly activated functional modules were identified (Figure 8B; stages Ta). One contains AKT serine/threonine kinase (AKT) from the phosphatidylinositol-4,5-bisphosphate 3-kinase (PI3K) pathway that has been shown to be deeply involved in bladder cancer [41]. Note that inhibitors targeting this protein have been recently developed [42] and show promising results. The OCSANA analysis suggests (Table 3) that Rho family of GTPases (RAS) de-phosphorylation should inhibit the propagation of the signal through this module to proliferation in Ta stage tumour. The second module contains mitogen-activated protein kinase 7 (MAP3K7) protein, an upstream activator of mitogen-activated protein kinase 14 (p38) and mitogen-activated protein kinase 8 (JNK), which is activated by three stimuli: TNF receptor superfamily member 1 (TNFR1), Interleukin 1 Receptor Type 1 (IL1R1) and Transforming Growth Factor Beta Receptor (TGFβR). All these pathways are frequently up-regulated in Ta tumours. OCSANA results suggested that the de-phosphorylation of both p38 and MAP3K7 can block the proliferative effects of MAP3K7-dependent functional module. Interestingly, bladder cancer cell lines were shown to proliferate because of the joint activity of PI3K and p38 (unpublished data), especially when FGFR3 is active. In the data set considered for the analysis, the FGFR3 gene was found to be strongly expressed in the stage Ta, less in the stage T1, whereas it has low expression in invasive bladder tumours. Strikingly, the best intervention for Ta tumours consists in the disruption of both p38 and a downstream target of PI3K, AKT that was identified as the most contributing gene with highest differential expression level in the path for the studied data set (Figure 8B). Table 3. Intervention sets for stage-specific activated modules in bladder cancer (adapted from [39]) Module Envisaged intervention (biochemical reactions) Ta modules  AKT_PHO RAS dephosphorylation  MAP3K7 p38 dephosphorylation; MAP3K7 de phosphorylation T1 modules  RE182 p70 knock-out; MYC knock-out  RE275 ERK dephosphorylation; RAS dephosphorylatio T2, T3, T4 modules  RE191 ERK dephosphorylation; RAS dephosphorylation  RE176  ATF2_PHO_JUN_PHO_ AT_NUCLEUS Module Envisaged intervention (biochemical reactions) Ta modules  AKT_PHO RAS dephosphorylation  MAP3K7 p38 dephosphorylation; MAP3K7 de phosphorylation T1 modules  RE182 p70 knock-out; MYC knock-out  RE275 ERK dephosphorylation; RAS dephosphorylatio T2, T3, T4 modules  RE191 ERK dephosphorylation; RAS dephosphorylation  RE176  ATF2_PHO_JUN_PHO_ AT_NUCLEUS Table 3. Intervention sets for stage-specific activated modules in bladder cancer (adapted from [39]) Module Envisaged intervention (biochemical reactions) Ta modules  AKT_PHO RAS dephosphorylation  MAP3K7 p38 dephosphorylation; MAP3K7 de phosphorylation T1 modules  RE182 p70 knock-out; MYC knock-out  RE275 ERK dephosphorylation; RAS dephosphorylatio T2, T3, T4 modules  RE191 ERK dephosphorylation; RAS dephosphorylation  RE176  ATF2_PHO_JUN_PHO_ AT_NUCLEUS Module Envisaged intervention (biochemical reactions) Ta modules  AKT_PHO RAS dephosphorylation  MAP3K7 p38 dephosphorylation; MAP3K7 de phosphorylation T1 modules  RE182 p70 knock-out; MYC knock-out  RE275 ERK dephosphorylation; RAS dephosphorylatio T2, T3, T4 modules  RE191 ERK dephosphorylation; RAS dephosphorylation  RE176  ATF2_PHO_JUN_PHO_ AT_NUCLEUS In the most invasive T2, T3 and T4 stages of bladder cancer, there are three up-regulated functional modules, all characterized by high expression of receptor tyrosine kinase/extracellular signal–regulated kinase (RTK/ERK) signalling components (Figure 8B; stages T2, T3 and T4). The OCSANA results (Table 3) point to the de-phosphorylation of both ERK and RAS as best interventions to block proliferation, which makes it coherent with the current developments of RAS- and ERK-inhibiting drugs for several cancer types, including bladder cancer[43]. The analysis suggested different interventions depending on the tumour stage. In less invasive tumours p38 coupled with PI3K-dependent signalling could be targeted, whereas RAS/ERK pathway is likely more critical in farther invasive stages. Similarly, network analysis using ACSN and OCSANA can be performed for individual patient tumour profiles, leading to personalized treatment recommendations [39]. Finding metastasis inducers in Colon cancer through network analysis Evolution of invasion and metastasis, in particular in colon cancer, has been extensively studied in experimental models. However, the mechanisms that trigger the process are still largely unknown, and the available mouse models of colon cancer are far from being satisfactory [44, 45]. To create an effective experimental mouse model of invasive colon cancer, it is fundamental to understand which are the major players, especially driver mutations, inducing invasion. As one of the early events of metastasis is assumed to be EMT [46], we started our exploration by focusing on this process. To identify the interplay between signalling pathways regulating EMT, a signalling network was manually created based on the information retrieved from around 200 publications. This signalling map is integrated into the EMT and cell motility comprehensive map of ACSN. Structural analysis and simplification of the EMT network highlighted the following EMT network organization principles, which are in agreement with current EMT understanding. (1) Five EMT transcription factors SNAIL, SLUG, TWIST, ZEB1 and ZEB2 have partially overlapping sets of downstream target genes that can activate the EMT-like program. (2) These key EMT transcription factors are under control of several upstream mechanisms: they are directly induced at the transcriptional level by the activated form of Notch, Notch Intracellular Domain (NICD) but are downregulated at the translational level by several miRNAs that are under transcriptional control of p53 family genes. (3) All five key EMT transcription factors should be activated ensuring simultaneous activation of EMT-like program genes and downregulating miRNAs. Additionally, the EMT key inducers also inhibit apoptosis and reduce proliferation. (4) The activity of the Wnt pathway is stimulated by the transcriptional activation of the gene coding for beta-catenin protein by NICD-induced TWIST or SNAI1. The Wnt pathway, in turn, can induce the expression of Notch pathway factors, creating a positive feedback loop. (5) Components of the Wnt and Notch pathways are negatively regulated by miRNAs induced by the p53 family (p53, p63 and p73). The balance between the effect of positive (Notch and Wnt) and negative (p53, p63 and p73 mediated by miRNAs) regulatory circuits on EMT inducers dictates the possibility of EMT phenotype [47–50]. Based on those features, the hub players were highlighted, and network complexity reduction was performed using the Cytoscape plugin BiNoM. The reduced network contained the core regulatory cascades of EMT, apoptosis and proliferation that were preserved through all levels of reduction [51]. This reduced network has been used for comparison between the wild type and all the possible combinations of single and double mutants that could promote an EMT phenotype. The computational analysis of the signalling network led to the prediction that the simultaneous activation of NICD and loss of p53 can promote an EMT phenotype. Furthermore, EMT inducers may activate the Wnt pathway, possibly resulting in a positive feedback loop that will amplify Notch activation and maintain an EMT-like program (Figure 9). Figure 9. View largeDownload slide Prediction of synthetic interaction combination to achieve EMT. Mechanistic model of EMT inducers regulation involving Notch (NICD), p53 and Wnt pathways in (A) normal and (B) double mutant with NICD overexpressed and p53 lost. (C) Scheme representing regulation of three major cell states in colon cancer (cell death, proliferation and metastasis) (Adapted from [52]). Figure 9. View largeDownload slide Prediction of synthetic interaction combination to achieve EMT. Mechanistic model of EMT inducers regulation involving Notch (NICD), p53 and Wnt pathways in (A) normal and (B) double mutant with NICD overexpressed and p53 lost. (C) Scheme representing regulation of three major cell states in colon cancer (cell death, proliferation and metastasis) (Adapted from [52]). To validate this hypothesis, a transgenic mouse model was generated, expressing a constitutively active Notch1 receptor in a p53-deleted background, specifically in the digestive epithelium. Importantly, green fluorescent protein expression linked to the Notch1 receptor activation allows lineage tracing of epithelial tumour cells during cancer progression and invasion (Figure 10A). These mice developed digestive tumours with dissemination of EMT-like epithelial malignant cells to the lymph nodes, liver and peritoneum, as well as generation of distant metastases (Figure 10B). Exploration of early EMT program inducers in invasive human colon cancer samples confirmed that EMT markers are associated with modulation of Notch and p53 gene expression in a similar manner as in the mouse model (Figure 10C), supporting a synergy between these genes to induce EMT [52]. Figure 10. View largeDownload slide p53 loss—Notch (NICD) overexpression double mutant results in invasive phenotype in colon cancer mice. Immunostaining for major EMT marker in (A) primary tumour and (B) metastases in distant organs; (C) regulation of p53, Notch and Wnt pathways in invasive colon cancer in human (TCGA data) (Adapted from [52]). Figure 10. View largeDownload slide p53 loss—Notch (NICD) overexpression double mutant results in invasive phenotype in colon cancer mice. Immunostaining for major EMT marker in (A) primary tumour and (B) metastases in distant organs; (C) regulation of p53, Notch and Wnt pathways in invasive colon cancer in human (TCGA data) (Adapted from [52]). The prediction of synthetic interaction between Notch (NICD) and p53 demonstrated that there are alternative ways to achieve conditions permissive of EMT, beyond those already described in the literature. This result was not obvious from the previous data and partially contradicts the commonly accepted dogma in the colon cancer field. The study supports an important message: gathering together cell signalling mechanisms may undercover unexpected interactions and lead to the discovery of new regulatory mechanisms of cell phenotypes that might significantly affect our understanding of basic molecular processes implicated in cancer, hence changing therapeutic approaches. In addition, the comprehensive EMT signalling network is a rich resource of information that can be used in further studies. Finally, the new EMT mice are a relevant model mimicking the invasive human colon cancer and a system for therapeutic drug discovery [53]. Studying heterogeneity of cancer-associated fibroblasts in breast cancer tumour microenvironment Carcinoma-associated fibroblasts (CAFs) are key players in the tumour microenvironment. They represent a heterogeneous population that can exhibit a range of polarization states from immune-stimulating to immune-suppressive, and therefore, impact in different ways tumour viability and development. To understand the subtle differences and the composition of CAF subpopulations in TNBC, signalling mechanisms responsible for various polarisation statuses of CAFs were gathered together into the cell-type-specific signalling map (a part of the immune response map of ACSN). Analysis of omics data from CAF subsets in TNBC patient performed in the context of the CAF map revealed that the two CAF subsets (CAF-S1, CAF-S4) accumulate differentially in TNBC patients and exhibit an opposite phenotype. Thus, CAF-S1 fibroblasts promote an immunosuppressive environment through a multistep mechanism and characterize patients with poor prognosis, whereas CAF-S4 fibroblasts are devoid of this immunosuppressive activity and therefore accumulate in patients with good prognosis [54]. Finding susceptibility to papillary thyroid carcinoma development The modules and maps of ACSN can serve as signatures of biological functions and can be used to find associations between perturbations in particular molecular mechanisms and the risk to develop a specific cancer type. In Lonjou etal., the association of 141 single nucleotide polymorphisms (SNPs) located in 43 DNA repair genes from 10 DNA repair processes as it is depicted in the DNA repair map, was examined in 75 papillary thyroid carcinoma (PTC) cases and 254 controls. The study confirms that genetic variants in several genes operating in distinct DNA repair mechanisms are implicated in the development of PTC. In particular, a significant association of the intronic SNP rs2296675 of the MGMT gene from the Direct Repair pathway with the risk of developing PTC was found [55]. Further investigation is underway, to decipher the molecular mechanisms controlled by the methyltransferase encoded by MGMT not only in the Direct Repair pathway but also in other associated mechanisms, with the aim of understanding how alteration of such functions may lead to the development of the most common type of thyroid cancer. Conclusions This review is devoted to the usage of signalling networks in cancer research, exemplified with the multidisciplinary project ACSN. It describes approaches addressing cancer complexity, by systematic representations of signalling pathways implicated in the disease in the form of comprehensive network maps and tools for network-based analysis, visualization and interpretation of cancer omics data. It summarized several studies applying ACSN to find synthetically interacting genes in cancer, predicting drug synergy, suggesting complex intervention sets, associating molecular mechanisms to cancer development susceptibility and the status of the tumor microenvironment. These examples of applications can serve as basis for a more general approach to understand signalling regulation in human disorders, to develop network-based models underlining drug resistance and to suggest intervention sets [1]. We depict a workflow that is applicable not only to cancer-related preclinical studies but also to the study other human diseases, such as central nervous system (CNS)-related, cardiovascular, metabolic and immune-related. With the aim of revealing perturbations of molecular mechanisms in different human diseases, to predict drug sensitivity, and to find optimal intervention schemes, one could combine the approaches represented in the examples previously described, into a workflow (Figure 11) that contains various steps: (i) construction of comprehensive intra- and intercellular signalling network of a disease; (ii) integration of omics data and retrieval of network-based signatures, characterizing the disease; (iii) on data availability, the resistance to treatment can be taken into account, and thus, mechanisms associated with the resistance can be highlighted in a form of deregulated functional modules, pathways or key players; (iv) modelling mechanisms governing the disease and drug resistance and computing intervention gene sets to interfere with the disease and drug resistance. The omics data from each patient can be taken into account, to rank the intervention gene sets according to intrinsic vulnerabilities in each patient. Figure 11. View largeDownload slide A workflow for studying human diseases using comprehensive cell signalling network maps together with omics data. Knowledge formalization in the form of comprehensive signalling network of intra- and intercellular molecular interactions followed by. Integration of multilevel omics data from patients and retrieval of network-based molecular portraits of the studied disease. Integration of drug resistance data (if applicable) and generation of weighted network-based signatures of drug resistance based on scoring of deregulated ‘biological functions’, pathways and key players. Extraction of deregulated sub-maps associated with the disease and/or drug resistance and modeling of drug resistance mechanisms. Structural analysis of deregulated sub-maps associated with drug resistance and finding intervention gene sets to interfere with cell survival and viability and restore sensitivity to drug. Scoring intervention sets using multi-omics data from each sample. Experimental validation of intervention sets in preclinical models and validation on patient data sets. Development of drug response predictor and set clinical trials (if applicable). Figure 11. View largeDownload slide A workflow for studying human diseases using comprehensive cell signalling network maps together with omics data. Knowledge formalization in the form of comprehensive signalling network of intra- and intercellular molecular interactions followed by. Integration of multilevel omics data from patients and retrieval of network-based molecular portraits of the studied disease. Integration of drug resistance data (if applicable) and generation of weighted network-based signatures of drug resistance based on scoring of deregulated ‘biological functions’, pathways and key players. Extraction of deregulated sub-maps associated with the disease and/or drug resistance and modeling of drug resistance mechanisms. Structural analysis of deregulated sub-maps associated with drug resistance and finding intervention gene sets to interfere with cell survival and viability and restore sensitivity to drug. Scoring intervention sets using multi-omics data from each sample. Experimental validation of intervention sets in preclinical models and validation on patient data sets. Development of drug response predictor and set clinical trials (if applicable). Finally, these steps should be followed by experimental validations in preclinical models, and once confirmed, validated in-patient studies. It will have an impact on personalized intervention schemes, in particular those based on pharmacological combination. A gradual integration of these approaches in the clinical routine will improve the response prediction to standard treatments, adjustment of intervention schemes and drug repositioning. The suggested approach demonstrated using ACSN has a wide potential and is applicable for other complex diseases; this rationale has led to the creation of a collective research effort on different human disorders called Disease Maps project (http://disease-maps.org). This partnership aims at applying similar approaches as described in this review that will lead to identifying emerging disease hallmarks of various disorders, as CNS-related Alzheimer’s [54] and Parkinson diseases [56], influenza [57] and others. This will help in studying disease comorbidities, predict response to standard treatments and to suggest improved individual intervention schemes based on drug repositioning [6]. In addition, there is an ongoing effort to include ACSN content into aggregated pathway databases, such as Pathway Commons [58] and WikiPathways [59] such that it can be used in many projects through standard interfaces as Cytoscape [60]. To study cross-regulation between cell signalling and metabolic processes, the integration of ACSN with the Virtual Metabolic Human (VMH) [61] currently takes place. Among others, this will allow to apply constrain-based modelling using the COBRA tool. This will provide an information about the biochemical reactions propagation taking place on the networks in two resources simultaneously [62]. Finally, to make the applications of ACSN broader, we ensure that the resource is compatible with, and well connected to downstream analysis pipeline tools. For example, ACSN is part of the GARUDA connectivity platform integrating inter-operable gadgets with applications in biology, healthcare and beyond [63] (http://www.garuda-alliance.org). Key Points ACSN is a resource of cancer signalling knowledge, comprehensive map of molecular interactions in cancer based on the latest scientific literature. NaviCell and NaviCom are interactive Web-based environments for molecular maps navigation, omics data integration and visualization. Interpretation of omics data from breast cancer samples with different sensitivity to targeted drugs in the context of signalling network helps to retrieve deregulated functional modules and to suggest synergy between drugs. A generalized network-based, data-driven approach for studying human disease is suggested. Luis Cristobal Monraz Gomez is a researcher at Institut Curie, Paris, France, his research is focused on a comprehensive representation of molecular mechanisms implicated in cancer, data analysis and interpretation in biomedical projects. Maria Kondratova is a researcher at Institut Curie, Paris, France, her scientific activity is focused on systematic visual representation of immune response mechanisms in cancer and study of cell reprogramming in tumor micronenvironment. Jean-Marie Ravel is resident in genetics at Nancy's university hospital medical genetic department. He is working as a collaborator on the Atlas of Cancer Signalling Networks (ACSN project). His research focuses on regulated cell death and angiogenesis signalling maps construction along with medical applications of this project. Emmanuel Barillot is the head of the Cancer and Genome: Bioinformatics, Biostatistics and Epidemiology of a Complex System department and scientific director of the bioinformatics platform at Institut Curie. His research focuses on methodological development and statistical analysis of high-throughput biological data and modeling with the aim to improve therapeutic treatments of cancer. Andrei Zinovyev is the scientific coordinator of the Computational Systems Biology of Cancer team at Institut Curie. His research focuses on high-throughput biological data analysis, complexity reduction, and modeling of biological networks involved in tumorigenesis and tumoral progression. Inna Kuperstein is a researcher at Institut Curie, Paris, France, she is a coordinator of the Atlas of Cancer Signalling Networks (ACSN) project for construction and analysis of detailed signalling maps, development of tools and modeling the maps to predict drug response. She participates in multidisciplinary projects to decipher cell mechanisms rewiring in cancer. Acknowledgements The authors thank Luca Albergante for critical reading of the article. Funding This work has been funded by the Agilent Thought Leader Award #3273 and ‘Projet Incitatif et Collaboratif Computational Systems Biology Approach for Cancer’ grant from Institut Curie. This work received support from APLIGOOGLE program provided by CNRS; the COLOSYS grant ANR-15-CMED-0001-04, provided by the Agence Nationale de la Recherche under the frame of ERACoSysMed-1, the ERA-Net for Systems Medicine in clinical research and medical practice and from INSERM Plan Cancer N° BIO2014-08 COMET grant under ITMO Cancer BioSys program. References 1 Barillot E , Calzone L , Hupe P , Vert J-PZA , Computational Systems Biology of Cancer . CRC Press . Boca Raton, FL , 2012 . 2 Carter H , Hofree M , Ideker T. Genotype to phenotype via network analysis . Curr Opin Genet Dev 2013 ; 23 ( 6 ): 611 – 21 . Google Scholar CrossRef Search ADS PubMed 3 Krogan NJ , Lippman S , Agard DA , et al. The cancer cell map initiative: defining the hallmark networks of cancer . Mol Cell 2015 ; 58 ( 4 ): 690 – 8 . Google Scholar CrossRef Search ADS PubMed 4 Chowdhury S , Sarkar RR. Comparison of human cell signaling pathway databases–evolution, drawbacks and challenges . Database 2015 ; 2015 : bau126 . Google Scholar CrossRef Search ADS PubMed 5 Wang J , Zuo Y , Man Y , et al. Pathway and network approaches for identification of cancer signature markers from omics data . J Cancer 2015 ; 6 ( 1 ): 54 – 65 . Google Scholar CrossRef Search ADS PubMed 6 Dorel M , Barillot E , Zinovyev A , et al. Network-based approaches for drug response prediction and targeted therapy development in cancer . Biochem Biophys Res Commun 2015 ; 464 ( 2 ): 386 – 91 . Google Scholar CrossRef Search ADS PubMed 7 Kaelin WG. The concept of synthetic lethality in the context of anticancer therapy . Nat Rev Cancer 2005 ; 5 ( 9 ): 689 – 98 . Google Scholar CrossRef Search ADS PubMed 8 Fang B. Development of synthetic lethality anticancer therapeutics . J Med Chem 2014 ; 57 ( 19 ): 7859 – 73 . Google Scholar CrossRef Search ADS PubMed 9 Garg A , Mohanram K , Di Cara A , et al. Efficient computation of minimal perturbation sets in gene regulatory networks . Front Physiol 2013 ; 4 : 361 . Google Scholar CrossRef Search ADS PubMed 10 Huang L , Li F , Sheng J , et al. DrugComboRanker: drug combination discovery based on target network analysis . Bioinformatics 2014 ; 30 ( 12 ): i228 – 36 . Google Scholar CrossRef Search ADS PubMed 11 Hanahan D , Weinberg RA. Hallmarks of cancer: the next generation . Cell 2011 ; 144 ( 5 ): 646 – 74 . Google Scholar CrossRef Search ADS PubMed 12 Kuperstein I , Bonnet E , Nguyen H-A , et al. Atlas of Cancer Signalling Network: a systems biology resource for integrative analysis of cancer data with Google Maps . Oncogenesis 2015 ; 4 : e160. Google Scholar CrossRef Search ADS PubMed 13 Kitano H , Funahashi A , Matsuoka Y , et al. Using process diagrams for the graphical representation of biological networks . Nat Biotechnol 2005 ; 23 : 961 – 6 . Google Scholar CrossRef Search ADS PubMed 14 Kuperstein I , Cohen DPA , Pook S , et al. NaviCell: a web-based environment for navigation, curation and maintenance of large molecular interaction maps . BMC Syst Biol 2013 ; 7 : 100 . Google Scholar CrossRef Search ADS PubMed 15 Khan FM , Marquardt S , Gupta SK , et al. Unraveling a tumor type-specific regulatory core underlying E2F1-mediated epithelial-mesenchymal transition to predict receptor protein signatures . Nat Commun 2017 ; 8 : 198 . Google Scholar CrossRef Search ADS PubMed 16 Niarakis A , Bounab Y , Grieco L , et al. Computational modeling of the main signaling pathways involved in mast cell activation . Curr Top Microbiol Immunol 2014 ; 382 : 69 – 93 . Google Scholar PubMed 17 Cantini L , Calzone L , Martignetti L , et al. Classification of gene signatures for their information value and functional redundancy . NPJ Syst Biol Appl 2018 ; 4 : 2 . Google Scholar CrossRef Search ADS PubMed 18 Bonnet E , Viara E , Kuperstein I , et al. NaviCell Web Service for network-based data visualization . Nucleic Acids Res 2015 ; 43 : W560 – 5 . Google Scholar CrossRef Search ADS PubMed 19 Dorel M , Viara E , Barillot E , et al. NaviCom: a web application to create interactive molecular network portraits using multi-level omics data . Database 2017 ; 2017 . doi: 10.1093/database/bax026. 20 Lapatas V , Stefanidakis M , Jimenez RC , et al. Data integration in biological research: an overview . J Biol Res 2015 ; 22 : 9 . 21 Kanehisa M , Goto S , Sato Y , et al. KEGG for integration and interpretation of large-scale molecular data sets . Nucleic Acids Res 2012 ; 40 : D109 – 14 . Google Scholar CrossRef Search ADS PubMed 22 Croft D , O’Kelly G , Wu G , et al. Reactome: a database of reactions, pathways and biological processes . Nucleic Acids Res 2010 ; 39 : D691 – 7 . Google Scholar CrossRef Search ADS PubMed 23 Subramanian A , Tamayo P , Mootha VK , et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles . Proc Natl Acad Sci USA 2005 ; 102 : 15545 – 50 . Google Scholar CrossRef Search ADS PubMed 24 Deveau P , Barillot E , Boeva V , et al. Calculating biological module enrichment or depletion and visualizing data on large-scale molecular maps with ACSNMineR and RNaviCell packages . R J 2016 ; 8 : 293 – 306 . 25 Martignetti L , Calzone L , Bonnet E , et al. ROMA: representation and quantification of module activity from target expression data . Front Genet 2016 ; 7 : 18. Google Scholar CrossRef Search ADS PubMed 26 Czerwinska U , Calzone L , Barillot E , et al. DeDaL: cytoscape 3 app for producing and morphing data-driven and structure-driven network layouts . BMC Syst Biol 2015 ; 9 : 46. Google Scholar CrossRef Search ADS PubMed 27 Lages J , Shepelyansky DL , Zinovyev A. Inferring hidden causal relations between pathway members using reduced Google matrix of directed biological networks . PLoS One 2018 ; 13 : e0190812 . doi: 10.1371/journal.pone.0190812. Google Scholar CrossRef Search ADS PubMed 28 Manié E , Popova T , Battistella A , et al. Genomic hallmarks of homologous recombination deficiency in invasive breast carcinomas . Int J cancer 2016 ; 138 : 891 – 900 . Google Scholar CrossRef Search ADS PubMed 29 Lord CJ , Tutt ANJ , Ashworth A. Synthetic lethality and cancer therapy: lessons learned from the development of PARP inhibitors . Annu Rev Med 2015 ; 66 : 455 – 70 . Google Scholar CrossRef Search ADS PubMed 30 Quanz M , Berthault N , Roulin C , et al. Small-molecule drugs mimicking DNA damage: a new strategy for sensitizing tumors to radiotherapy . Clin Cancer Res 2009 ; 15 ( 4 ): 1308 – 16 . Google Scholar CrossRef Search ADS PubMed 31 Jdey W , Thierry S , Russo C , et al. Drug-driven synthetic lethality: bypassing tumor cell genetics with a combination of AsiDNA and PARP inhibitors . Clin Cancer Res 2017 ; 23 ( 4 ): 1001 – 11 . Google Scholar CrossRef Search ADS PubMed 32 McLornan DP , List A , Mufti GJ. Applying synthetic lethality for the selective targeting of cancer . N Engl J Med 2014 ; 371 ( 18 ): 1725 – 35 . Google Scholar CrossRef Search ADS PubMed 33 Dietlein F , Thelen L , Reinhardt HC. Cancer-specific defects in DNA repair pathways as targets for personalized therapeutic approaches . Trends Genet 2014 ; 30 : 326 – 39 . Google Scholar CrossRef Search ADS PubMed 34 Acencio ML , Bovolenta LA , Camilo E , et al. Prediction of oncogenic interactions and cancer-related signaling networks based on network topology . PLoS One 2013 ; 8 ( 10 ): e77521 . Google Scholar CrossRef Search ADS PubMed 35 Zeng T , Wang DC , Wang X , et al. Prediction of dynamical drug sensitivity and resistance by module network rewiring-analysis based on transcriptional profiling . Drug Resist Updat 2014 ; 17 : 64 – 76 . Google Scholar CrossRef Search ADS PubMed 36 Grieco L , Calzone L , Bernard-Pierrot I , et al. Integrative modelling of the influence of MAPK network on cancer cell fate decision . PLoS Comput Biol 2013 ; 9 ( 10 ): e1003286 . Google Scholar CrossRef Search ADS PubMed 37 Stransky N , Vallot C , Reyal F , et al. Regional copy number-independent deregulation of transcription in cancer . Nat Genet 2006 ; 38 : 1386 – 96 . Google Scholar CrossRef Search ADS PubMed 38 Kuperstein I , Grieco L , Cohen DPA , et al. The shortest path is not the one you know: application of biological network resources in precision oncology research . Mutagenesis 2015 ; 30 ( 2 ): 191 – 204 . Google Scholar CrossRef Search ADS PubMed 39 Grieco L. Integrative modelling and analysis of MAPK network deregulations in human cancers. PhD thesis, Aix-Marseille University, 2013 , 1 – 51 . 40 Vera-Licona P , Bonnet E , Barillot E , et al. OCSANA: optimal combinations of interventions from network analysis . Bioinformatics 2013 ; 29 : 1571 – 3 . Google Scholar CrossRef Search ADS PubMed 41 Askham JM , Platt F , Chambers PA , et al. AKT1 mutations in bladder cancer: identification of a novel oncogenic mutation that can co-operate with E17K . Oncogene 2010 ; 29 : 150 – 5 . Google Scholar CrossRef Search ADS PubMed 42 Ching CB , Hansel DE. Expanding therapeutic targets in bladder cancer: the PI3K/Akt/mTOR pathway . Lab Invest 2010 ; 90 : 1406 – 14 . Google Scholar CrossRef Search ADS PubMed 43 Cakir M , Grossman AB. Targeting MAPK (Ras/ERK) and PI3K/Akt pathways in pituitary tumorigenesis . Expert Opin Ther Targets 2009 ; 13 ( 9 ): 1121 – 34 . Google Scholar CrossRef Search ADS PubMed 44 Hung KE , Maricevich MA , Richard LG , et al. Development of a mouse model for sporadic and metastatic colon tumors and its use in assessing drug treatment . Proc Natl Acad Sci USA 2010 ; 107 ( 4 ): 1565 – 70 . Google Scholar CrossRef Search ADS PubMed 45 Trobridge P , Knoblaugh S , Washington MK , et al. TGF-beta receptor inactivation and mutant Kras induce intestinal neoplasms in mice via a beta-catenin-independent pathway . Gastroenterology 2009 ; 136 : 1680 – 8.e7 . Google Scholar CrossRef Search ADS PubMed 46 Nieto MA. The ins and outs of the epithelial to mesenchymal transition in health and disease . Annu Rev Cell Dev Biol 2011 ; 27 : 347 – 76 . Google Scholar CrossRef Search ADS PubMed 47 Knouf EC , Garg K , Arroyo JD , et al. An integrative genomic approach identifies p73 and p63 as activators of miR-200 microRNA family transcription . Nucleic Acids Res 2012 ; 40 ( 2 ): 499 – 510 . Google Scholar CrossRef Search ADS PubMed 48 Moes M , Le Béchec A , Crespo I , et al. A novel network integrating a miRNA-203/SNAI1 feedback loop which regulates epithelial to mesenchymal transition . PLoS One 2012 ; 7 ( 4 ): e35440 . Google Scholar CrossRef Search ADS PubMed 49 Siemens H , Jackstadt R , Hünten S , et al. miR-34 and SNAIL form a double-negative feedback loop to regulate epithelial-mesenchymal transitions . Cell Cycle 2011 ; 10 : 4256 – 71 . Google Scholar CrossRef Search ADS PubMed 50 Fre S , Pallavi SK , Huyghe M , et al. Notch and Wnt signals cooperatively control cell proliferation and tumorigenesis in the intestine . Proc Natl Acad Sci 2009 ; 106 ( 15 ): 6309 – 14 . Google Scholar CrossRef Search ADS PubMed 51 Bonnet E , Calzone L , Rovera D , et al. BiNoM 2.0, a Cytoscape plugin for accessing and analyzing pathways using standard systems biology formats . BMC Syst Biol 2013 ; 7 : 18 . Google Scholar CrossRef Search ADS PubMed 52 Chanrion M , Kuperstein I , Barrière C , et al. Concomitant Notch activation and p53 deletion trigger epithelial-to-mesenchymal transition and metastasis in mouse gut . Nat Commun 2014 ; 5 : 5005 . Google Scholar CrossRef Search ADS PubMed 53 Kuperstein I , Robine S , Zinovyev A. Network biology elucidates metastatic colon cancer mechanisms . Cell Cycle 2015 ; 14 ( 14 ): 2189 – 90 . Google Scholar CrossRef Search ADS PubMed 54 Costa A , Kieffer Y , Scholer-Dahirel A , et al. Fibroblast heterogeneity and immunosuppressive environment in human breast cancer . Cancer Cell 2018 ; 33 ( 3 ): 463 – 79.e10 . Google Scholar CrossRef Search ADS PubMed 55 Lonjou C , Damiola F , Moissonnier M , et al. Investigation of DNA repair-related SNPs underlying susceptibility to papillary thyroid carcinoma reveals MGMT as a novel candidate gene in Belarusian children exposed to radiation . BMC Cancer 2017 ; 17 : 1 – 9 . Google Scholar CrossRef Search ADS PubMed 56 Fujita KA , Ostaszewski M , Matsuoka Y , et al. Integrating pathways of Parkinson’s disease in a molecular interaction map . Mol Neurobiol 2014 ; 49 ( 1 ): 88 – 102 . Google Scholar CrossRef Search ADS PubMed 57 Matsuoka Y , Matsumae H , Katoh M , et al. A comprehensive map of the influenza A virus replication cycle . BMC Syst Biol 2013 ; 7 : 97. Google Scholar CrossRef Search ADS PubMed 58 Cerami EG , Gross BE , Demir E , et al. Pathway Commons, a web resource for biological pathway data . Nucleic Acids Res 2011 ; 39 : D685 – 90 . Google Scholar CrossRef Search ADS PubMed 59 Slenter DN , Kutmon M , Hanspers K , et al. WikiPathways: a multifaceted pathway database bridging metabolomics to other omics research . Nucleic Acids Res 2018 ; 46 : D661 – 7 . Google Scholar CrossRef Search ADS PubMed 60 Shannon P , Markiel A , Ozier O , et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks . Genome Res 2003 ; 13 : 2498 – 504 . Google Scholar CrossRef Search ADS PubMed 61 Noronha A , Daníelsdóttir AD , Gawron P , et al. ReconMap: an interactive visualization of human metabolism . Bioinformatics 2016 ; 33 : 605 – 7 . 62 Heirendt L , Arreckx S , Pfau T , et al. Creation and analysis of biochemical constraint-based models: the COBRA Toolbox v3.0 . arXiv 2017 ;1710.04038. 63 Ghosh S , Matsuoka Y , Asai Y , et al. Toward an integrated software platform for systems pharmacology . Biopharm Drug Dispos 2013 ; 34 : 508 – 26 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Briefings in Bioinformatics Oxford University Press

Application of Atlas of Cancer Signalling Network in preclinical studies

Loading next page...
 
/lp/ou_press/application-of-atlas-of-cancer-signalling-network-in-preclinical-md5Mc9SJJO
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com
ISSN
1467-5463
eISSN
1477-4054
D.O.I.
10.1093/bib/bby031
Publisher site
See Article on Publisher Site

Abstract

Abstract Cancer initiation and progression are associated with multiple molecular mechanisms. The knowledge of these mechanisms is expanding and should be converted into guidelines for tackling the disease. Here, we discuss the formalization of biological knowledge into a comprehensive resource: the Atlas of Cancer Signalling Network (ACSN) and the Google Maps-based tool NaviCell, which supports map navigation. The application of ACSN for omics data visualization, in the context of signalling maps, is possible via the NaviCell Web Service module and through the NaviCom tool. It allows generation of network-based molecular portraits of cancer using multilevel omics data. We review how these resources and tools are applied for cancer preclinical studies. Structural analysis of the maps together with omics data helps to rationalize the synergistic effects of drugs and allows design of complex disease stage-specific druggable interventions. The use of ACSN modules and maps as signatures of biological functions can help in cancer data analysis and interpretation. In addition, they empowered finding of associations between perturbations in particular molecular mechanisms and the risk to develop a specific type of cancer. These approaches are helpful, among others, to study the interplay between molecular mechanisms of cancer. It opens an opportunity to decipher how gene interactions govern the hallmarks of cancer in specific contexts. We discuss a perspective to develop a flexible methodology and a pipeline to enable systematic omics data analysis in the context of signalling network maps, for stratifying patients and suggesting interventions points and drug repositioning in cancer and other diseases. systems biology, cancer, signalling network maps, omics data integration and analysis, data visualization, synthetic interactions, drug response Introduction According to the current scientific understanding, different signalling pathways interact to create a complex network featuring feed forward loops, backward regulatory loops and many alternative paths that support redundancy. It is generally believed that, under most pathological transformations, cells do not exploit new molecular signalling mechanisms but rather hijack existing molecular programs. This affects not only intracellular functions but also the interactions between different cell types, leading to a new, yet pathological, status of the system. It is likely that a specific combination of molecular characteristics dictates specific cell signalling states, maintaining the pathological disease status. Identifying and manipulating the key molecular players that control these cell signalling states, and shifting the pathological cell state towards the desired healthy or, alternatively, lethal phenotype, are major challenges for molecular biology both in the more general context of human diseases and in the specific context of cancer [1–3]. To allow an appropriate data analysis of the molecular mechanisms supporting carcinogenesis, the information on these mechanisms should be systematically and adequately represented. The knowledge concerning the molecular signalling mechanisms of the cell is spread across thousands of publications, mostly in a human-readable, and computer-unfriendly, form, precluding the direct application of many bioinformatics and systems biology approaches. Hence, it is important to compile such knowledge in a computer-readable form. The most common way to approach this problem is to represent the relationships between molecules in a network form, resulting in pathway diagrams, which can be found in various pathway databases [4]. As the amount of information about biological mechanisms steadily increases, introducing a different approach to organize and structure this information is essential. Our aim is to provide a global cell signalling picture with sufficient granularity to preserve molecular details, capturing crosstalks and feedback loops between molecular circuits. For this purpose, comprehensive signalling network maps covering simultaneously multiple cellular processes are more suitable than disconnected pathway diagrams. This review focuses on different approaches for a rational representation of cell signalling in cancer, and will use the Atlas of Cancer Signalling Network (ACSN) as an example. Visualization and analysis of omics data in the context of signalling networks facilitate data interpretation and can help in highlighting deregulated mechanisms. Furthermore, data analysis in the context of signalling networks can help detecting patterns in the data projected onto the molecular mechanisms represented in the signalling maps, providing information on enriched functional modules (‘hot’ deregulated areas), key players and ‘bottleneck’ points [3, 5]. Correlating the status of those network variables with the phenotype, as drug resistance or patient survival, can be followed by clustering approaches to allow us stratifying patients according to their network-based molecular portraits. Moreover, these approaches can help in suggesting intervention points and in designing appropriate therapeutic schemes [6]. As a productive idea for intervention schemes, synthetic lethality (SL) provides a conceptual framework to develop cancer-specific drugs. This classical paradigm defines SL interactions as a phenomenon where combinations of two gene deletions significantly compromise cell viability, whereas the deletion of one of those genes does not [7]. The idea of the SL treatment approach is to take advantage of the vulnerabilities in tumour cells, which can be characterized by the abnormal function of one of the genes from the SL pair. Targeting the SL partner allows then to selectively kill the tumour cells, avoiding or reducing side effects on normal cells [8]. The approach also provides the clinician with a biomarker to select patients that could respond to the treatment. Giving the complexity of signalling mechanisms simultaneously involved in cancer, the SL pair paradigm should be extended to the SL sets or combinations paradigm [9, 10]. The computational approaches that allow in silico testing of multiple synthetic interactions combinations, considering large comprehensive signalling networks and cancer omics data, are discussed in this review. ACSN: geographical map of molecular mechanisms Deregulation of molecular mechanisms leading to cancer can be observed in various processes such as cell cycle, cell death, DNA repair and replication, cell motility and adhesion, cell survival mechanisms, immune processes, angiogenesis, tumour microenvironment and others. Most of them are collectively or sequentially involved in tumour formation and modified as the tumour evolves. The scientific literature often suggests that in pathological situations, the normal cell signalling network is altered by deregulated coordination between pathways or disruption of existing molecular pathways, rather than by creating completely new signalling pathways. The most common abnormalities in pathological situations are perturbations at the gene expression level, protein abundance or protein post-translational modifications, irregular ‘firing’ or silencing of particular signals, wrong subcellular localization of particular molecules and so on. Such quantitative rather than qualitative network changes, compared with the normal cell signalling, could be studied in the context of comprehensive signalling networks by analysing experimental data obtained from tumour samples, patient-derived xenografts, cancer-related cell lines or animal models. This approach helps to understand the interplay between molecular mechanisms in cancer and to decipher how gene and protein interactions govern the hallmarks of cancer [11], in specific settings. Despite the existence of a large variety of pathway databases and resources [4], only few of them are cancer-specific, and none of these resources depict the processes with sufficient granularity. In addition, pathway browsing interfaces are becoming more important for cancer researchers and clinicians but require further improvements. Therefore, we constructed a resource, the ACSN (http://acsn.curie.fr) that aims at formalizing the knowledge on cancer-related processes in the form of a comprehensive signalling network map, for data interpretation in basic research and preclinical studies [12]. The construction and update of ACSN involve manual mining of the cellular and molecular biology literature, along with the participation of experts in those fields. ACSN differs from other databases because it contains a deep comprehensive description of cancer-related mechanisms, retrieved from the most recent literature, based on the hallmarks of cancer (Figure 1). Cell signalling mechanisms are depicted using the CellDesigner tool [13] at the level of biochemical interactions, assembling a large network of 4826 reactions covering 2371 proteins and based on approximately 3000 references (Table 1). Table 1. Content of ACSN (adapted from [12]) Feature Content Maps of biological processes 5 Functional modules 52 Chemical species 5975 Reactions 4826 Proteins 2371 Metabolites 595 Genes 159 References 2919 Feature Content Maps of biological processes 5 Functional modules 52 Chemical species 5975 Reactions 4826 Proteins 2371 Metabolites 595 Genes 159 References 2919 Table 1. Content of ACSN (adapted from [12]) Feature Content Maps of biological processes 5 Functional modules 52 Chemical species 5975 Reactions 4826 Proteins 2371 Metabolites 595 Genes 159 References 2919 Feature Content Maps of biological processes 5 Functional modules 52 Chemical species 5975 Reactions 4826 Proteins 2371 Metabolites 595 Genes 159 References 2919 Figure 1. View largeDownload slide Structure of ACSN resource. The scheme demonstrates the concept of ACSN construction starting from the cancer hallmarks: collecting information about molecular mechanisms underlying those hallmarks from scientific publications and manually depicting them in the global map of ACSN and further supporting by consulting the information from the external pathway databases. ACSN is hierarchically organized into three levels: the seamless global map divided into the interconnected biological process maps that are further decomposed into interconnected module maps. ACSN can be exploited through Web-based NaviCell interface allowing map navigation using Google Maps engine, map commenting via associated blog system and user omics data visualization and analysis (Adapted from [12]). Figure 1. View largeDownload slide Structure of ACSN resource. The scheme demonstrates the concept of ACSN construction starting from the cancer hallmarks: collecting information about molecular mechanisms underlying those hallmarks from scientific publications and manually depicting them in the global map of ACSN and further supporting by consulting the information from the external pathway databases. ACSN is hierarchically organized into three levels: the seamless global map divided into the interconnected biological process maps that are further decomposed into interconnected module maps. ACSN can be exploited through Web-based NaviCell interface allowing map navigation using Google Maps engine, map commenting via associated blog system and user omics data visualization and analysis (Adapted from [12]). Currently, ACSN contains representations of molecular mechanisms that are frequently dysregulated in cancer, such as cell cycle, DNA repair, cell death, cell survival and epithelial to mesenchymal transition (EMT). Cell signalling mechanisms are depicted in the maps in great detail, creating together a map of molecular interactions, presented as a global ‘geographic-like’ representation (Figure 1A). ACSN has a hierarchical structure, composed of interconnected maps of processes altered in cancer. Each map is further divided into functional modules, corresponding mainly to canonical signalling pathways (Figures 1 and 2C). Figure 2. View largeDownload slide Browsing interface of ACSN. (A) ACSN interface with selection panel and data visualization menu. Querying ACSN is possible via the search window or by checking on the entity in the list of entities. Distribution of frequently mutated oncogene Myc proto-oncogene protein (MYC) across molecular mechanisms on the ACSN maps is indicated; (B) Google Maps-like features of NaviCell for visualization and annotation of map entities (markers, call-out with links to external databases, citations and ACSN maps of and functional modules where MYC protein is found); (C) Zoom in on Wnt non-canonical module of cell survival map to observe signalling processes where MYC protein is involved (Adapted from[12]). Figure 2. View largeDownload slide Browsing interface of ACSN. (A) ACSN interface with selection panel and data visualization menu. Querying ACSN is possible via the search window or by checking on the entity in the list of entities. Distribution of frequently mutated oncogene Myc proto-oncogene protein (MYC) across molecular mechanisms on the ACSN maps is indicated; (B) Google Maps-like features of NaviCell for visualization and annotation of map entities (markers, call-out with links to external databases, citations and ACSN maps of and functional modules where MYC protein is found); (C) Zoom in on Wnt non-canonical module of cell survival map to observe signalling processes where MYC protein is involved (Adapted from[12]). The navigation interface includes features such as scrolling, zooming, markers and callouts using the Google Maps technology adapted by NaviCell [14] (Figure 3), Web-based platform supporting ACSN and similar efforts in CellDesigner format [15, 16] or other formats [17]. The semantic zooming in NaviCell (http://navicell.curie.fr) provides several view levels, achieved by gradual exclusion of details and abstraction of information on zooming out (Figure 2B). Figure 3. View largeDownload slide General architecture of NaviCell environment (Adapted from [18]). Figure 3. View largeDownload slide General architecture of NaviCell environment (Adapted from [18]). ACSN is a unique resource of cancer signalling knowledge, with an enormous amount of information embedded and organized. Together with NaviCell, it is optimized for integration and visualization of cancer molecular profiles generated by high-throughput technologies, drug screening data or synthetic interactions studies. The integration and analysis of these data in the context of ACSN may help to better understand the biological relevance of results, guiding scientific hypotheses and suggesting potential therapeutic intervention for cancer patients. In addition, as ACSN covers major cell signalling processes, the resource and associated methods for data analysis using ACSN are suitable for applications in many biological fields and for studying various human diseases. The Atlas is currently being extended with additional maps depicting molecular mechanisms of DNA replication, telomere maintenance, angiogenesis, immune response and others that will be integrated in future releases. The Atlas will not only cover intracellular processes but also crosstalks of cancer cell with the components of tumour microenvironment. An additional level of complexity will be added to the Atlas in the near future, representing different types of cells surrounding the tumour, and their interplay, to enable modelling of complex phenotypes. Molecular portraits of cancer: data visualization and analysis using ACSN Data visualization in NaviCell Web Service environment The data integration into ACSN is possible through NaviCell Web Service, a user-friendly interface making a part of the NaviCell tool [18]. It allows uploading several types of ‘omics’ data—e.g. mRNA expression data, microRNA, proteins, mutation profiles and copy number data—in simple text table (tab-separated) format. The data can be visualized simultaneously in the context of molecular interaction maps (Figure 4). Detailed instructions, tutorials and life examples of multi-omics data visualization using NaviCell Web Service are available at https://navicell.curie.fr/pages/nav_web_service.html. Figure 4. View largeDownload slide General architecture of NaviCell Web service server. Client software (light blue layer) communicates with the server (red layer) through standard HTTP requests using the standard JSON format to encode data (RESTful Web service, dark blue layer). A session (with a unique ID) is established between the server and the browser (yellow layer) through Ajax communication channel to visualize the results of the commands send by the software client (Adapted from [18]). Figure 4. View largeDownload slide General architecture of NaviCell Web service server. Client software (light blue layer) communicates with the server (red layer) through standard HTTP requests using the standard JSON format to encode data (RESTful Web service, dark blue layer). A session (with a unique ID) is established between the server and the browser (yellow layer) through Ajax communication channel to visualize the results of the commands send by the software client (Adapted from [18]). Depending on the type of data, different visualization modes can be applied to obtain an informative picture, such as bar plots, glyphs and map staining (Table 2). The data can be visualized at different zoom levels. Sample annotation files uploaded with the data can serve to define groups of samples. Table 2. Data display modes in NaviCell and NaviCom (adapted from [19]) Data type Visualization on mode Data display Units mRNA expression Map staining Level Gene copy number Heat map Count Muta on data Glyph 1 Frequency Methyla on data Glyph 2 Intensity miRNA expression Glyph 3 Level Protein expression Glyph 4 Level Data type Visualization on mode Data display Units mRNA expression Map staining Level Gene copy number Heat map Count Muta on data Glyph 1 Frequency Methyla on data Glyph 2 Intensity miRNA expression Glyph 3 Level Protein expression Glyph 4 Level Table 2. Data display modes in NaviCell and NaviCom (adapted from [19]) Data type Visualization on mode Data display Units mRNA expression Map staining Level Gene copy number Heat map Count Muta on data Glyph 1 Frequency Methyla on data Glyph 2 Intensity miRNA expression Glyph 3 Level Protein expression Glyph 4 Level Data type Visualization on mode Data display Units mRNA expression Map staining Level Gene copy number Heat map Count Muta on data Glyph 1 Frequency Methyla on data Glyph 2 Intensity miRNA expression Glyph 3 Level Protein expression Glyph 4 Level A novel mode of data visualization for continuous data (e.g. expression) provided by the NaviCell Web Service is the ‘map staining’. With this technique, the values mapped to individual molecular entities or group of entities (e.g. score of functional module activities, see below) result in a colourful background of the network map [18]. All the approaches for data integration into the signalling maps described above allow to elucidate and interpret the omics data, compare samples or groups, find patterns across the molecular mechanisms depicted on the maps, grasp deregulated ‘hot areas’ on the maps and major involved players and draw hypotheses as to which mechanisms to focus on. These signalling network-based molecular signatures thus help to stratify patients (Figure 5). Figure 5. View largeDownload slide Breast cancer gene expression data integration and analysis using NaviCell. The mRNA expression data from TCGA collection has been used for evaluation of functional modules activities and ACSN colouring as ‘map staining’ for (A) basal-like and (B) luminal A breast cancer types. The two breast cancer subtypes are characterized by different patterns of module activities. (Adapted from [12]). Figure 5. View largeDownload slide Breast cancer gene expression data integration and analysis using NaviCell. The mRNA expression data from TCGA collection has been used for evaluation of functional modules activities and ACSN colouring as ‘map staining’ for (A) basal-like and (B) luminal A breast cancer types. The two breast cancer subtypes are characterized by different patterns of module activities. (Adapted from [12]). Various omics data are available in the public and commercial repositories [20]. However, there is a lack of tools supporting the integration of big data sets from these databases and the visualization on signalling network maps in an efficient way and with optimized visualization settings. To answer to this demand, NaviCom has been developed, a python package providing a Web interface for simultaneous display of multilevel data in the context of signalling network maps [19]. NaviCom (http://navicom.curie.fr) provides a bridge between the cBioPortal database and the NaviCell interactive tool for data visualization. NaviCom integrates functionalities from the cBioFetchR R package to import high-throughput data sets from cBioPortal to NaviCell and the Navicom Python module, allowing automatized simultaneous visualization of multilevel omics data on the interactive signalling network maps provided by the NaviCell environment (Figure 6A). NaviCom proposes several standardized modes of data visualization on signalling network maps to address specific biological questions (Table 2). Figure 6. View largeDownload slide General architecture of NaviCom environment. The NaviCom interface provides the user with an updated list of studies from cBioPortal and links to ACSN and NaviCell maps collections. When visualization is launched, NaviCom starts a new NaviCell session and calls a cgi on the server. The cgi downloads cBioPortal data to the NaviCell session and displays them to generate the molecular portrait selected by the user (Adapted from [19]). Figure 6. View largeDownload slide General architecture of NaviCom environment. The NaviCom interface provides the user with an updated list of studies from cBioPortal and links to ACSN and NaviCell maps collections. When visualization is launched, NaviCom starts a new NaviCell session and calls a cgi on the server. The cgi downloads cBioPortal data to the NaviCell session and displays them to generate the molecular portrait selected by the user (Adapted from [19]). This tool enables the generation of complex molecular portraits from multiple omics data sets from cBioPortal. Detailed instructions and tutorials are available at https://navicom.curie.fr/tutorial.ph. In the near future, the NaviCom platform will be extended and will provide access to many types of omics data from a wide range of databases (TCGA, ICGC, HGMB, METABRIC and CCLE). In addition, to allow for a wider description of the molecular mechanisms implicated in the studied sample, signalling networks available in databases as KEGG [21], Reactome [22] and others, will be also integrated and used for high-throughput data analysis via the NaviCom platform. Data analysis using ACSN To identify dysregulated signalling pathways of functional modules in a molecular map from a given data set, several tools are currently available. Gene Set Enrichment Analysis (GSEA) (http://software.broadinstitute.org/gsea/index.jsp) is a computational method aimed at finding overrepresented modules in a ranked gene list, using a weighted Kolmogorov–Smirnov test [23]. ACSNMineR (https://github.com/sysbio-curie/ACSNMineR) is an R package that incorporates ACSN information to calculate enriched or depleted modules by means of a Fisher exact test or a hypergeometric test [24]. The Representation and Quantification of Module Activity (ROMA) method, implemented in Java and R (https://github.com/sysbio-curie/Roma, https://github.com/sysbio-curie/rRoma), is designed for fast and robust computation of the activity of gene sets (or modules) with coordinated expression. ROMA uses the first component of principal component analysis to summarize the co-expression of a group of genes in a gene set. ROMA also proves additional functionalities: (i) calculation of the individual gene contribution to the module activity level and determination of the genes that are contributing the most to the first principal component, (ii) several variants of computation for the first principal component, i.e. weighted and centred methods and (iii) estimation of the statistical significance of the proportion of variance explained by the first principal component, as well as the spectral gap between the variance explained by the first and second component (representing the homogeneity for the gene set) [25]. The module activity scores calculated by these methods can be visualized in the context of ACSN using the map staining technique as described above (Figure 5). Such visualization is automated in ACSNMineR and ROMA. ACSN was also used as a source of module definitions for benchmarking the DeDaL tool. DeDaL allows the creation of a mixed data-driven and structure-driven network layout, which can be more insightful for grasping the correlation patterns in the multivariate data on top of the networks [26]. ACSN module definitions were applied for testing a method for inferring hidden causal relations between pathway members using reduced Google matrix of directed biological networks [27]. ACSN acts as a source of functional module definitions and protein–protein interactions network in data analysis projects, especially those from cancer data. For example, gene lists from functional modules of the DNA repair map were used to study homologous recombination (HR) deficiency in invasive breast carcinomas [28]. ACSN was used as a source of signatures for processes involved in cancer for classification of gene signatures and generation of InfoSigMap, an interactive online map showing the structure of compositional and functional redundancies between signatures from various sources [17]. Exploiting the ACSN in preclinical research Explaining the synergistic effect of combined treatments in breast cancer DNA repair inhibitors are holding promises to improve cancer therapy, but their application is limited by the compensatory activities of different repair pathways in cancer cells. For example, PARP inhibitors, which act as SL with BRCA deficiency, appear less efficient in patients with active HR repair mechanisms [29]. During treatment, some tumours escape the elimination through compensatory mutations that restore the HR activity or stimulate the activity of alternative repair pathways such as non-homologous end joining (NHEJ) and alternative non-homologous end joining (Alt-NHEJ). A new class of DNA repair pathways inhibitors (Dbait or AsiDNA, a Dbait derivative) has been recently developed, consisting of 32 bp deoxyribonucleotides DNA double helix that mimics double-strand breaks (DSB). It acts as an agonist of DNA damage signalling, thereby inhibiting DNA repair enzyme recruitment at the damage site [30]. However, studies on the effects of Dbait on multiple types of cancer cell lines show occurrences of resistance in a cancer type-independent manner. Depending on the genetic background, different breast cancer tumours vary in their sensitivity to DNA repair inhibitors, as PARP inhibitors and AsiDNA. To understand the molecular mechanisms underlining these differences, a combination of experimental and bioinformatics approaches was applied. Triple negative breast cancer (TNBC) cell lines were studied for their sensitivity to AsiDNA, the derivative of Dbait DNA repair inhibitor, and Olaparib, the PARP inhibitor. Different TNBC cell lines show a wide distribution of response/resistance to these drugs, despite the fact that these cell lines are related to the same disease type. Integrative analyses of omics data from these cell lines encompassing mRNA expression, copy number variations and mutational profiles were performed, retrieving non-overlapping unique gene sets robustly correlated with resistance to each one of the drugs. Analysis of the omics data in the context of ACSN maps highlighted dysregulated functional modules across ACSN, associated with resistance to each one of the drugs allowing to establish drug resistance network-based molecular portraits. This analysis confirmed that different specific defects in DNA repair machinery are associated to AsiDNA (Figure 7A) or Olaparib (Figure 7B) resistance. Importantly, it showed involvement of different compensatory DNA repair mechanisms in cell lines resistant to AsiDNA when compared with cell lines resistant to Olaparib (Figure 7D), suggesting a rationale to combine these two drugs. The authors confirmed a synergistic therapeutic effect of the combined treatment with AsiDNA and PARP inhibitors in TNBC, while sparing healthy tissue (Figure 7C) [31]. Figure 7. View largeDownload slide Overcoming resistance of TNBC cell lines to DNA repair inhibitors. Molecular portraits of TNBC cell lines resistant to (A) AsiDNA or (B) Olaparib, visualized on DNA repair map. (C) Cell survival to combination of AsiDNA and Olaparib, with AsiDNA 1(black line), without AsiDNA (grey line); dashed lines indicate calculated cell survival for additive effect of two drugs. (D) Schematic representation of inhibitory mechanisms of AsiDNA and Olaparib. Base excision repair (BER), HR, NHEJ, Alt-NHEJ (Adapted from [31]). Figure 7. View largeDownload slide Overcoming resistance of TNBC cell lines to DNA repair inhibitors. Molecular portraits of TNBC cell lines resistant to (A) AsiDNA or (B) Olaparib, visualized on DNA repair map. (C) Cell survival to combination of AsiDNA and Olaparib, with AsiDNA 1(black line), without AsiDNA (grey line); dashed lines indicate calculated cell survival for additive effect of two drugs. (D) Schematic representation of inhibitory mechanisms of AsiDNA and Olaparib. Base excision repair (BER), HR, NHEJ, Alt-NHEJ (Adapted from [31]). Complex stage-specific interventions in MAPK pathway to disrupt proliferative signalling in bladder cancer The idea of an SL treatment approach is to take advantage of the peculiarities of tumour cells having an abnormal expression or functionality of one gene from an SL pair. Targeting another SL partner allows selective killing of the tumour cells [32]. This approach is applied in BRCA2 mutated breast cancer cases using PARP inhibitors. However, there is a frequent escape from the treatment, requiring a more complex approach. Treatment failure can be because of the robustness of cell signalling network ensured by redundant mechanisms that provide the possibility to bypass the effect of drugs [33]. Therefore, the ways for identifying and blocking those active compensatory pathways should be found. One of the approaches to solve this issue is by taking into account the signalling network structure to find the most optimal SL gene combinations, possibly more than a pair [10, 34, 35]. A computational strategy to suggest complex intervention sets has been developed and demonstrated using the mitogen-activated protein kinase (MAPK) signalling network [36]. The MAPK signalling network is coordinated with various processes implicated in cell survival, and currently included into the Cell Survival map of ACSN. The strategy involves two steps: (i) identification of tumour stage-specific active functional modules, i.e. sets of MAPK signalling network components that are transcriptionally deregulated in bladder cancer [37] compared with normal samples, and (ii) computation of intervention sets of MAPK map components, whose disruption block all the proliferative paths fostered by the identified active functional modules in bladder cancer [38] (Figure 8A). Figure 8. View largeDownload slide Computational strategy for finding stage-specific interventions sets using detailed reaction network analysis and omics data. (A) Detailed MAPK network map is shown with schematically indicated activated modules (see text). Finding minimal hitting sets allows to cut all paths (schematically shown by arrows) from the activated modules to the proliferation phenotype. (B) Stage-specific activated modules detected in MAPK network using bladder cancer transcriptome data. Module enrichment score were computed by GSEA method. The most contributing leading-edge genes with highest differential expression level are highlighted by red. The optimal hitting set lists from these gene elements of MAPK network, which on removal cuts all the paths from the corresponding activated modules to the proliferation phenotype, were calculated using the OCSANA algorithm, and intervention sets for each stage of bladder cancer were suggested (see text) (Adapted from [39]). Figure 8. View largeDownload slide Computational strategy for finding stage-specific interventions sets using detailed reaction network analysis and omics data. (A) Detailed MAPK network map is shown with schematically indicated activated modules (see text). Finding minimal hitting sets allows to cut all paths (schematically shown by arrows) from the activated modules to the proliferation phenotype. (B) Stage-specific activated modules detected in MAPK network using bladder cancer transcriptome data. Module enrichment score were computed by GSEA method. The most contributing leading-edge genes with highest differential expression level are highlighted by red. The optimal hitting set lists from these gene elements of MAPK network, which on removal cuts all the paths from the corresponding activated modules to the proliferation phenotype, were calculated using the OCSANA algorithm, and intervention sets for each stage of bladder cancer were suggested (see text) (Adapted from [39]). The procedure was applied to five different bladder cancer stages [37]. Differential gene expression levels were computed, relative to healthy conditions, using transcriptomic data. Regions of the map having high density of differentially expressed genes/proteins were identified and scored by GSEA [23]. The highly scoring regions were assumed to likely point to sources of proliferative signals in the tumour. If paths exist in the map from the components belonging to any strongly activated network region to the node ‘Proliferation’, then presumably the region contributes to the activation of cell division (Figure 8A). The removal of a set of proteins from the network might block all such proliferative paths. This idea was formalized by the notion of the minimal cut sets, which were computed using the OCSANA algorithm in the Cytoscape plugin BiNoM [40]. The OCSANA algorithm computes the minimal cut sets, by simultaneously prioritizing them with respect to the potential effect on the target network nodes, while avoiding side effects on the parts of the network that functionally should be preserved. In the less invasive Ta stage, two significantly activated functional modules were identified (Figure 8B; stages Ta). One contains AKT serine/threonine kinase (AKT) from the phosphatidylinositol-4,5-bisphosphate 3-kinase (PI3K) pathway that has been shown to be deeply involved in bladder cancer [41]. Note that inhibitors targeting this protein have been recently developed [42] and show promising results. The OCSANA analysis suggests (Table 3) that Rho family of GTPases (RAS) de-phosphorylation should inhibit the propagation of the signal through this module to proliferation in Ta stage tumour. The second module contains mitogen-activated protein kinase 7 (MAP3K7) protein, an upstream activator of mitogen-activated protein kinase 14 (p38) and mitogen-activated protein kinase 8 (JNK), which is activated by three stimuli: TNF receptor superfamily member 1 (TNFR1), Interleukin 1 Receptor Type 1 (IL1R1) and Transforming Growth Factor Beta Receptor (TGFβR). All these pathways are frequently up-regulated in Ta tumours. OCSANA results suggested that the de-phosphorylation of both p38 and MAP3K7 can block the proliferative effects of MAP3K7-dependent functional module. Interestingly, bladder cancer cell lines were shown to proliferate because of the joint activity of PI3K and p38 (unpublished data), especially when FGFR3 is active. In the data set considered for the analysis, the FGFR3 gene was found to be strongly expressed in the stage Ta, less in the stage T1, whereas it has low expression in invasive bladder tumours. Strikingly, the best intervention for Ta tumours consists in the disruption of both p38 and a downstream target of PI3K, AKT that was identified as the most contributing gene with highest differential expression level in the path for the studied data set (Figure 8B). Table 3. Intervention sets for stage-specific activated modules in bladder cancer (adapted from [39]) Module Envisaged intervention (biochemical reactions) Ta modules  AKT_PHO RAS dephosphorylation  MAP3K7 p38 dephosphorylation; MAP3K7 de phosphorylation T1 modules  RE182 p70 knock-out; MYC knock-out  RE275 ERK dephosphorylation; RAS dephosphorylatio T2, T3, T4 modules  RE191 ERK dephosphorylation; RAS dephosphorylation  RE176  ATF2_PHO_JUN_PHO_ AT_NUCLEUS Module Envisaged intervention (biochemical reactions) Ta modules  AKT_PHO RAS dephosphorylation  MAP3K7 p38 dephosphorylation; MAP3K7 de phosphorylation T1 modules  RE182 p70 knock-out; MYC knock-out  RE275 ERK dephosphorylation; RAS dephosphorylatio T2, T3, T4 modules  RE191 ERK dephosphorylation; RAS dephosphorylation  RE176  ATF2_PHO_JUN_PHO_ AT_NUCLEUS Table 3. Intervention sets for stage-specific activated modules in bladder cancer (adapted from [39]) Module Envisaged intervention (biochemical reactions) Ta modules  AKT_PHO RAS dephosphorylation  MAP3K7 p38 dephosphorylation; MAP3K7 de phosphorylation T1 modules  RE182 p70 knock-out; MYC knock-out  RE275 ERK dephosphorylation; RAS dephosphorylatio T2, T3, T4 modules  RE191 ERK dephosphorylation; RAS dephosphorylation  RE176  ATF2_PHO_JUN_PHO_ AT_NUCLEUS Module Envisaged intervention (biochemical reactions) Ta modules  AKT_PHO RAS dephosphorylation  MAP3K7 p38 dephosphorylation; MAP3K7 de phosphorylation T1 modules  RE182 p70 knock-out; MYC knock-out  RE275 ERK dephosphorylation; RAS dephosphorylatio T2, T3, T4 modules  RE191 ERK dephosphorylation; RAS dephosphorylation  RE176  ATF2_PHO_JUN_PHO_ AT_NUCLEUS In the most invasive T2, T3 and T4 stages of bladder cancer, there are three up-regulated functional modules, all characterized by high expression of receptor tyrosine kinase/extracellular signal–regulated kinase (RTK/ERK) signalling components (Figure 8B; stages T2, T3 and T4). The OCSANA results (Table 3) point to the de-phosphorylation of both ERK and RAS as best interventions to block proliferation, which makes it coherent with the current developments of RAS- and ERK-inhibiting drugs for several cancer types, including bladder cancer[43]. The analysis suggested different interventions depending on the tumour stage. In less invasive tumours p38 coupled with PI3K-dependent signalling could be targeted, whereas RAS/ERK pathway is likely more critical in farther invasive stages. Similarly, network analysis using ACSN and OCSANA can be performed for individual patient tumour profiles, leading to personalized treatment recommendations [39]. Finding metastasis inducers in Colon cancer through network analysis Evolution of invasion and metastasis, in particular in colon cancer, has been extensively studied in experimental models. However, the mechanisms that trigger the process are still largely unknown, and the available mouse models of colon cancer are far from being satisfactory [44, 45]. To create an effective experimental mouse model of invasive colon cancer, it is fundamental to understand which are the major players, especially driver mutations, inducing invasion. As one of the early events of metastasis is assumed to be EMT [46], we started our exploration by focusing on this process. To identify the interplay between signalling pathways regulating EMT, a signalling network was manually created based on the information retrieved from around 200 publications. This signalling map is integrated into the EMT and cell motility comprehensive map of ACSN. Structural analysis and simplification of the EMT network highlighted the following EMT network organization principles, which are in agreement with current EMT understanding. (1) Five EMT transcription factors SNAIL, SLUG, TWIST, ZEB1 and ZEB2 have partially overlapping sets of downstream target genes that can activate the EMT-like program. (2) These key EMT transcription factors are under control of several upstream mechanisms: they are directly induced at the transcriptional level by the activated form of Notch, Notch Intracellular Domain (NICD) but are downregulated at the translational level by several miRNAs that are under transcriptional control of p53 family genes. (3) All five key EMT transcription factors should be activated ensuring simultaneous activation of EMT-like program genes and downregulating miRNAs. Additionally, the EMT key inducers also inhibit apoptosis and reduce proliferation. (4) The activity of the Wnt pathway is stimulated by the transcriptional activation of the gene coding for beta-catenin protein by NICD-induced TWIST or SNAI1. The Wnt pathway, in turn, can induce the expression of Notch pathway factors, creating a positive feedback loop. (5) Components of the Wnt and Notch pathways are negatively regulated by miRNAs induced by the p53 family (p53, p63 and p73). The balance between the effect of positive (Notch and Wnt) and negative (p53, p63 and p73 mediated by miRNAs) regulatory circuits on EMT inducers dictates the possibility of EMT phenotype [47–50]. Based on those features, the hub players were highlighted, and network complexity reduction was performed using the Cytoscape plugin BiNoM. The reduced network contained the core regulatory cascades of EMT, apoptosis and proliferation that were preserved through all levels of reduction [51]. This reduced network has been used for comparison between the wild type and all the possible combinations of single and double mutants that could promote an EMT phenotype. The computational analysis of the signalling network led to the prediction that the simultaneous activation of NICD and loss of p53 can promote an EMT phenotype. Furthermore, EMT inducers may activate the Wnt pathway, possibly resulting in a positive feedback loop that will amplify Notch activation and maintain an EMT-like program (Figure 9). Figure 9. View largeDownload slide Prediction of synthetic interaction combination to achieve EMT. Mechanistic model of EMT inducers regulation involving Notch (NICD), p53 and Wnt pathways in (A) normal and (B) double mutant with NICD overexpressed and p53 lost. (C) Scheme representing regulation of three major cell states in colon cancer (cell death, proliferation and metastasis) (Adapted from [52]). Figure 9. View largeDownload slide Prediction of synthetic interaction combination to achieve EMT. Mechanistic model of EMT inducers regulation involving Notch (NICD), p53 and Wnt pathways in (A) normal and (B) double mutant with NICD overexpressed and p53 lost. (C) Scheme representing regulation of three major cell states in colon cancer (cell death, proliferation and metastasis) (Adapted from [52]). To validate this hypothesis, a transgenic mouse model was generated, expressing a constitutively active Notch1 receptor in a p53-deleted background, specifically in the digestive epithelium. Importantly, green fluorescent protein expression linked to the Notch1 receptor activation allows lineage tracing of epithelial tumour cells during cancer progression and invasion (Figure 10A). These mice developed digestive tumours with dissemination of EMT-like epithelial malignant cells to the lymph nodes, liver and peritoneum, as well as generation of distant metastases (Figure 10B). Exploration of early EMT program inducers in invasive human colon cancer samples confirmed that EMT markers are associated with modulation of Notch and p53 gene expression in a similar manner as in the mouse model (Figure 10C), supporting a synergy between these genes to induce EMT [52]. Figure 10. View largeDownload slide p53 loss—Notch (NICD) overexpression double mutant results in invasive phenotype in colon cancer mice. Immunostaining for major EMT marker in (A) primary tumour and (B) metastases in distant organs; (C) regulation of p53, Notch and Wnt pathways in invasive colon cancer in human (TCGA data) (Adapted from [52]). Figure 10. View largeDownload slide p53 loss—Notch (NICD) overexpression double mutant results in invasive phenotype in colon cancer mice. Immunostaining for major EMT marker in (A) primary tumour and (B) metastases in distant organs; (C) regulation of p53, Notch and Wnt pathways in invasive colon cancer in human (TCGA data) (Adapted from [52]). The prediction of synthetic interaction between Notch (NICD) and p53 demonstrated that there are alternative ways to achieve conditions permissive of EMT, beyond those already described in the literature. This result was not obvious from the previous data and partially contradicts the commonly accepted dogma in the colon cancer field. The study supports an important message: gathering together cell signalling mechanisms may undercover unexpected interactions and lead to the discovery of new regulatory mechanisms of cell phenotypes that might significantly affect our understanding of basic molecular processes implicated in cancer, hence changing therapeutic approaches. In addition, the comprehensive EMT signalling network is a rich resource of information that can be used in further studies. Finally, the new EMT mice are a relevant model mimicking the invasive human colon cancer and a system for therapeutic drug discovery [53]. Studying heterogeneity of cancer-associated fibroblasts in breast cancer tumour microenvironment Carcinoma-associated fibroblasts (CAFs) are key players in the tumour microenvironment. They represent a heterogeneous population that can exhibit a range of polarization states from immune-stimulating to immune-suppressive, and therefore, impact in different ways tumour viability and development. To understand the subtle differences and the composition of CAF subpopulations in TNBC, signalling mechanisms responsible for various polarisation statuses of CAFs were gathered together into the cell-type-specific signalling map (a part of the immune response map of ACSN). Analysis of omics data from CAF subsets in TNBC patient performed in the context of the CAF map revealed that the two CAF subsets (CAF-S1, CAF-S4) accumulate differentially in TNBC patients and exhibit an opposite phenotype. Thus, CAF-S1 fibroblasts promote an immunosuppressive environment through a multistep mechanism and characterize patients with poor prognosis, whereas CAF-S4 fibroblasts are devoid of this immunosuppressive activity and therefore accumulate in patients with good prognosis [54]. Finding susceptibility to papillary thyroid carcinoma development The modules and maps of ACSN can serve as signatures of biological functions and can be used to find associations between perturbations in particular molecular mechanisms and the risk to develop a specific cancer type. In Lonjou etal., the association of 141 single nucleotide polymorphisms (SNPs) located in 43 DNA repair genes from 10 DNA repair processes as it is depicted in the DNA repair map, was examined in 75 papillary thyroid carcinoma (PTC) cases and 254 controls. The study confirms that genetic variants in several genes operating in distinct DNA repair mechanisms are implicated in the development of PTC. In particular, a significant association of the intronic SNP rs2296675 of the MGMT gene from the Direct Repair pathway with the risk of developing PTC was found [55]. Further investigation is underway, to decipher the molecular mechanisms controlled by the methyltransferase encoded by MGMT not only in the Direct Repair pathway but also in other associated mechanisms, with the aim of understanding how alteration of such functions may lead to the development of the most common type of thyroid cancer. Conclusions This review is devoted to the usage of signalling networks in cancer research, exemplified with the multidisciplinary project ACSN. It describes approaches addressing cancer complexity, by systematic representations of signalling pathways implicated in the disease in the form of comprehensive network maps and tools for network-based analysis, visualization and interpretation of cancer omics data. It summarized several studies applying ACSN to find synthetically interacting genes in cancer, predicting drug synergy, suggesting complex intervention sets, associating molecular mechanisms to cancer development susceptibility and the status of the tumor microenvironment. These examples of applications can serve as basis for a more general approach to understand signalling regulation in human disorders, to develop network-based models underlining drug resistance and to suggest intervention sets [1]. We depict a workflow that is applicable not only to cancer-related preclinical studies but also to the study other human diseases, such as central nervous system (CNS)-related, cardiovascular, metabolic and immune-related. With the aim of revealing perturbations of molecular mechanisms in different human diseases, to predict drug sensitivity, and to find optimal intervention schemes, one could combine the approaches represented in the examples previously described, into a workflow (Figure 11) that contains various steps: (i) construction of comprehensive intra- and intercellular signalling network of a disease; (ii) integration of omics data and retrieval of network-based signatures, characterizing the disease; (iii) on data availability, the resistance to treatment can be taken into account, and thus, mechanisms associated with the resistance can be highlighted in a form of deregulated functional modules, pathways or key players; (iv) modelling mechanisms governing the disease and drug resistance and computing intervention gene sets to interfere with the disease and drug resistance. The omics data from each patient can be taken into account, to rank the intervention gene sets according to intrinsic vulnerabilities in each patient. Figure 11. View largeDownload slide A workflow for studying human diseases using comprehensive cell signalling network maps together with omics data. Knowledge formalization in the form of comprehensive signalling network of intra- and intercellular molecular interactions followed by. Integration of multilevel omics data from patients and retrieval of network-based molecular portraits of the studied disease. Integration of drug resistance data (if applicable) and generation of weighted network-based signatures of drug resistance based on scoring of deregulated ‘biological functions’, pathways and key players. Extraction of deregulated sub-maps associated with the disease and/or drug resistance and modeling of drug resistance mechanisms. Structural analysis of deregulated sub-maps associated with drug resistance and finding intervention gene sets to interfere with cell survival and viability and restore sensitivity to drug. Scoring intervention sets using multi-omics data from each sample. Experimental validation of intervention sets in preclinical models and validation on patient data sets. Development of drug response predictor and set clinical trials (if applicable). Figure 11. View largeDownload slide A workflow for studying human diseases using comprehensive cell signalling network maps together with omics data. Knowledge formalization in the form of comprehensive signalling network of intra- and intercellular molecular interactions followed by. Integration of multilevel omics data from patients and retrieval of network-based molecular portraits of the studied disease. Integration of drug resistance data (if applicable) and generation of weighted network-based signatures of drug resistance based on scoring of deregulated ‘biological functions’, pathways and key players. Extraction of deregulated sub-maps associated with the disease and/or drug resistance and modeling of drug resistance mechanisms. Structural analysis of deregulated sub-maps associated with drug resistance and finding intervention gene sets to interfere with cell survival and viability and restore sensitivity to drug. Scoring intervention sets using multi-omics data from each sample. Experimental validation of intervention sets in preclinical models and validation on patient data sets. Development of drug response predictor and set clinical trials (if applicable). Finally, these steps should be followed by experimental validations in preclinical models, and once confirmed, validated in-patient studies. It will have an impact on personalized intervention schemes, in particular those based on pharmacological combination. A gradual integration of these approaches in the clinical routine will improve the response prediction to standard treatments, adjustment of intervention schemes and drug repositioning. The suggested approach demonstrated using ACSN has a wide potential and is applicable for other complex diseases; this rationale has led to the creation of a collective research effort on different human disorders called Disease Maps project (http://disease-maps.org). This partnership aims at applying similar approaches as described in this review that will lead to identifying emerging disease hallmarks of various disorders, as CNS-related Alzheimer’s [54] and Parkinson diseases [56], influenza [57] and others. This will help in studying disease comorbidities, predict response to standard treatments and to suggest improved individual intervention schemes based on drug repositioning [6]. In addition, there is an ongoing effort to include ACSN content into aggregated pathway databases, such as Pathway Commons [58] and WikiPathways [59] such that it can be used in many projects through standard interfaces as Cytoscape [60]. To study cross-regulation between cell signalling and metabolic processes, the integration of ACSN with the Virtual Metabolic Human (VMH) [61] currently takes place. Among others, this will allow to apply constrain-based modelling using the COBRA tool. This will provide an information about the biochemical reactions propagation taking place on the networks in two resources simultaneously [62]. Finally, to make the applications of ACSN broader, we ensure that the resource is compatible with, and well connected to downstream analysis pipeline tools. For example, ACSN is part of the GARUDA connectivity platform integrating inter-operable gadgets with applications in biology, healthcare and beyond [63] (http://www.garuda-alliance.org). Key Points ACSN is a resource of cancer signalling knowledge, comprehensive map of molecular interactions in cancer based on the latest scientific literature. NaviCell and NaviCom are interactive Web-based environments for molecular maps navigation, omics data integration and visualization. Interpretation of omics data from breast cancer samples with different sensitivity to targeted drugs in the context of signalling network helps to retrieve deregulated functional modules and to suggest synergy between drugs. A generalized network-based, data-driven approach for studying human disease is suggested. Luis Cristobal Monraz Gomez is a researcher at Institut Curie, Paris, France, his research is focused on a comprehensive representation of molecular mechanisms implicated in cancer, data analysis and interpretation in biomedical projects. Maria Kondratova is a researcher at Institut Curie, Paris, France, her scientific activity is focused on systematic visual representation of immune response mechanisms in cancer and study of cell reprogramming in tumor micronenvironment. Jean-Marie Ravel is resident in genetics at Nancy's university hospital medical genetic department. He is working as a collaborator on the Atlas of Cancer Signalling Networks (ACSN project). His research focuses on regulated cell death and angiogenesis signalling maps construction along with medical applications of this project. Emmanuel Barillot is the head of the Cancer and Genome: Bioinformatics, Biostatistics and Epidemiology of a Complex System department and scientific director of the bioinformatics platform at Institut Curie. His research focuses on methodological development and statistical analysis of high-throughput biological data and modeling with the aim to improve therapeutic treatments of cancer. Andrei Zinovyev is the scientific coordinator of the Computational Systems Biology of Cancer team at Institut Curie. His research focuses on high-throughput biological data analysis, complexity reduction, and modeling of biological networks involved in tumorigenesis and tumoral progression. Inna Kuperstein is a researcher at Institut Curie, Paris, France, she is a coordinator of the Atlas of Cancer Signalling Networks (ACSN) project for construction and analysis of detailed signalling maps, development of tools and modeling the maps to predict drug response. She participates in multidisciplinary projects to decipher cell mechanisms rewiring in cancer. Acknowledgements The authors thank Luca Albergante for critical reading of the article. Funding This work has been funded by the Agilent Thought Leader Award #3273 and ‘Projet Incitatif et Collaboratif Computational Systems Biology Approach for Cancer’ grant from Institut Curie. This work received support from APLIGOOGLE program provided by CNRS; the COLOSYS grant ANR-15-CMED-0001-04, provided by the Agence Nationale de la Recherche under the frame of ERACoSysMed-1, the ERA-Net for Systems Medicine in clinical research and medical practice and from INSERM Plan Cancer N° BIO2014-08 COMET grant under ITMO Cancer BioSys program. References 1 Barillot E , Calzone L , Hupe P , Vert J-PZA , Computational Systems Biology of Cancer . CRC Press . Boca Raton, FL , 2012 . 2 Carter H , Hofree M , Ideker T. Genotype to phenotype via network analysis . Curr Opin Genet Dev 2013 ; 23 ( 6 ): 611 – 21 . Google Scholar CrossRef Search ADS PubMed 3 Krogan NJ , Lippman S , Agard DA , et al. The cancer cell map initiative: defining the hallmark networks of cancer . Mol Cell 2015 ; 58 ( 4 ): 690 – 8 . Google Scholar CrossRef Search ADS PubMed 4 Chowdhury S , Sarkar RR. Comparison of human cell signaling pathway databases–evolution, drawbacks and challenges . Database 2015 ; 2015 : bau126 . Google Scholar CrossRef Search ADS PubMed 5 Wang J , Zuo Y , Man Y , et al. Pathway and network approaches for identification of cancer signature markers from omics data . J Cancer 2015 ; 6 ( 1 ): 54 – 65 . Google Scholar CrossRef Search ADS PubMed 6 Dorel M , Barillot E , Zinovyev A , et al. Network-based approaches for drug response prediction and targeted therapy development in cancer . Biochem Biophys Res Commun 2015 ; 464 ( 2 ): 386 – 91 . Google Scholar CrossRef Search ADS PubMed 7 Kaelin WG. The concept of synthetic lethality in the context of anticancer therapy . Nat Rev Cancer 2005 ; 5 ( 9 ): 689 – 98 . Google Scholar CrossRef Search ADS PubMed 8 Fang B. Development of synthetic lethality anticancer therapeutics . J Med Chem 2014 ; 57 ( 19 ): 7859 – 73 . Google Scholar CrossRef Search ADS PubMed 9 Garg A , Mohanram K , Di Cara A , et al. Efficient computation of minimal perturbation sets in gene regulatory networks . Front Physiol 2013 ; 4 : 361 . Google Scholar CrossRef Search ADS PubMed 10 Huang L , Li F , Sheng J , et al. DrugComboRanker: drug combination discovery based on target network analysis . Bioinformatics 2014 ; 30 ( 12 ): i228 – 36 . Google Scholar CrossRef Search ADS PubMed 11 Hanahan D , Weinberg RA. Hallmarks of cancer: the next generation . Cell 2011 ; 144 ( 5 ): 646 – 74 . Google Scholar CrossRef Search ADS PubMed 12 Kuperstein I , Bonnet E , Nguyen H-A , et al. Atlas of Cancer Signalling Network: a systems biology resource for integrative analysis of cancer data with Google Maps . Oncogenesis 2015 ; 4 : e160. Google Scholar CrossRef Search ADS PubMed 13 Kitano H , Funahashi A , Matsuoka Y , et al. Using process diagrams for the graphical representation of biological networks . Nat Biotechnol 2005 ; 23 : 961 – 6 . Google Scholar CrossRef Search ADS PubMed 14 Kuperstein I , Cohen DPA , Pook S , et al. NaviCell: a web-based environment for navigation, curation and maintenance of large molecular interaction maps . BMC Syst Biol 2013 ; 7 : 100 . Google Scholar CrossRef Search ADS PubMed 15 Khan FM , Marquardt S , Gupta SK , et al. Unraveling a tumor type-specific regulatory core underlying E2F1-mediated epithelial-mesenchymal transition to predict receptor protein signatures . Nat Commun 2017 ; 8 : 198 . Google Scholar CrossRef Search ADS PubMed 16 Niarakis A , Bounab Y , Grieco L , et al. Computational modeling of the main signaling pathways involved in mast cell activation . Curr Top Microbiol Immunol 2014 ; 382 : 69 – 93 . Google Scholar PubMed 17 Cantini L , Calzone L , Martignetti L , et al. Classification of gene signatures for their information value and functional redundancy . NPJ Syst Biol Appl 2018 ; 4 : 2 . Google Scholar CrossRef Search ADS PubMed 18 Bonnet E , Viara E , Kuperstein I , et al. NaviCell Web Service for network-based data visualization . Nucleic Acids Res 2015 ; 43 : W560 – 5 . Google Scholar CrossRef Search ADS PubMed 19 Dorel M , Viara E , Barillot E , et al. NaviCom: a web application to create interactive molecular network portraits using multi-level omics data . Database 2017 ; 2017 . doi: 10.1093/database/bax026. 20 Lapatas V , Stefanidakis M , Jimenez RC , et al. Data integration in biological research: an overview . J Biol Res 2015 ; 22 : 9 . 21 Kanehisa M , Goto S , Sato Y , et al. KEGG for integration and interpretation of large-scale molecular data sets . Nucleic Acids Res 2012 ; 40 : D109 – 14 . Google Scholar CrossRef Search ADS PubMed 22 Croft D , O’Kelly G , Wu G , et al. Reactome: a database of reactions, pathways and biological processes . Nucleic Acids Res 2010 ; 39 : D691 – 7 . Google Scholar CrossRef Search ADS PubMed 23 Subramanian A , Tamayo P , Mootha VK , et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles . Proc Natl Acad Sci USA 2005 ; 102 : 15545 – 50 . Google Scholar CrossRef Search ADS PubMed 24 Deveau P , Barillot E , Boeva V , et al. Calculating biological module enrichment or depletion and visualizing data on large-scale molecular maps with ACSNMineR and RNaviCell packages . R J 2016 ; 8 : 293 – 306 . 25 Martignetti L , Calzone L , Bonnet E , et al. ROMA: representation and quantification of module activity from target expression data . Front Genet 2016 ; 7 : 18. Google Scholar CrossRef Search ADS PubMed 26 Czerwinska U , Calzone L , Barillot E , et al. DeDaL: cytoscape 3 app for producing and morphing data-driven and structure-driven network layouts . BMC Syst Biol 2015 ; 9 : 46. Google Scholar CrossRef Search ADS PubMed 27 Lages J , Shepelyansky DL , Zinovyev A. Inferring hidden causal relations between pathway members using reduced Google matrix of directed biological networks . PLoS One 2018 ; 13 : e0190812 . doi: 10.1371/journal.pone.0190812. Google Scholar CrossRef Search ADS PubMed 28 Manié E , Popova T , Battistella A , et al. Genomic hallmarks of homologous recombination deficiency in invasive breast carcinomas . Int J cancer 2016 ; 138 : 891 – 900 . Google Scholar CrossRef Search ADS PubMed 29 Lord CJ , Tutt ANJ , Ashworth A. Synthetic lethality and cancer therapy: lessons learned from the development of PARP inhibitors . Annu Rev Med 2015 ; 66 : 455 – 70 . Google Scholar CrossRef Search ADS PubMed 30 Quanz M , Berthault N , Roulin C , et al. Small-molecule drugs mimicking DNA damage: a new strategy for sensitizing tumors to radiotherapy . Clin Cancer Res 2009 ; 15 ( 4 ): 1308 – 16 . Google Scholar CrossRef Search ADS PubMed 31 Jdey W , Thierry S , Russo C , et al. Drug-driven synthetic lethality: bypassing tumor cell genetics with a combination of AsiDNA and PARP inhibitors . Clin Cancer Res 2017 ; 23 ( 4 ): 1001 – 11 . Google Scholar CrossRef Search ADS PubMed 32 McLornan DP , List A , Mufti GJ. Applying synthetic lethality for the selective targeting of cancer . N Engl J Med 2014 ; 371 ( 18 ): 1725 – 35 . Google Scholar CrossRef Search ADS PubMed 33 Dietlein F , Thelen L , Reinhardt HC. Cancer-specific defects in DNA repair pathways as targets for personalized therapeutic approaches . Trends Genet 2014 ; 30 : 326 – 39 . Google Scholar CrossRef Search ADS PubMed 34 Acencio ML , Bovolenta LA , Camilo E , et al. Prediction of oncogenic interactions and cancer-related signaling networks based on network topology . PLoS One 2013 ; 8 ( 10 ): e77521 . Google Scholar CrossRef Search ADS PubMed 35 Zeng T , Wang DC , Wang X , et al. Prediction of dynamical drug sensitivity and resistance by module network rewiring-analysis based on transcriptional profiling . Drug Resist Updat 2014 ; 17 : 64 – 76 . Google Scholar CrossRef Search ADS PubMed 36 Grieco L , Calzone L , Bernard-Pierrot I , et al. Integrative modelling of the influence of MAPK network on cancer cell fate decision . PLoS Comput Biol 2013 ; 9 ( 10 ): e1003286 . Google Scholar CrossRef Search ADS PubMed 37 Stransky N , Vallot C , Reyal F , et al. Regional copy number-independent deregulation of transcription in cancer . Nat Genet 2006 ; 38 : 1386 – 96 . Google Scholar CrossRef Search ADS PubMed 38 Kuperstein I , Grieco L , Cohen DPA , et al. The shortest path is not the one you know: application of biological network resources in precision oncology research . Mutagenesis 2015 ; 30 ( 2 ): 191 – 204 . Google Scholar CrossRef Search ADS PubMed 39 Grieco L. Integrative modelling and analysis of MAPK network deregulations in human cancers. PhD thesis, Aix-Marseille University, 2013 , 1 – 51 . 40 Vera-Licona P , Bonnet E , Barillot E , et al. OCSANA: optimal combinations of interventions from network analysis . Bioinformatics 2013 ; 29 : 1571 – 3 . Google Scholar CrossRef Search ADS PubMed 41 Askham JM , Platt F , Chambers PA , et al. AKT1 mutations in bladder cancer: identification of a novel oncogenic mutation that can co-operate with E17K . Oncogene 2010 ; 29 : 150 – 5 . Google Scholar CrossRef Search ADS PubMed 42 Ching CB , Hansel DE. Expanding therapeutic targets in bladder cancer: the PI3K/Akt/mTOR pathway . Lab Invest 2010 ; 90 : 1406 – 14 . Google Scholar CrossRef Search ADS PubMed 43 Cakir M , Grossman AB. Targeting MAPK (Ras/ERK) and PI3K/Akt pathways in pituitary tumorigenesis . Expert Opin Ther Targets 2009 ; 13 ( 9 ): 1121 – 34 . Google Scholar CrossRef Search ADS PubMed 44 Hung KE , Maricevich MA , Richard LG , et al. Development of a mouse model for sporadic and metastatic colon tumors and its use in assessing drug treatment . Proc Natl Acad Sci USA 2010 ; 107 ( 4 ): 1565 – 70 . Google Scholar CrossRef Search ADS PubMed 45 Trobridge P , Knoblaugh S , Washington MK , et al. TGF-beta receptor inactivation and mutant Kras induce intestinal neoplasms in mice via a beta-catenin-independent pathway . Gastroenterology 2009 ; 136 : 1680 – 8.e7 . Google Scholar CrossRef Search ADS PubMed 46 Nieto MA. The ins and outs of the epithelial to mesenchymal transition in health and disease . Annu Rev Cell Dev Biol 2011 ; 27 : 347 – 76 . Google Scholar CrossRef Search ADS PubMed 47 Knouf EC , Garg K , Arroyo JD , et al. An integrative genomic approach identifies p73 and p63 as activators of miR-200 microRNA family transcription . Nucleic Acids Res 2012 ; 40 ( 2 ): 499 – 510 . Google Scholar CrossRef Search ADS PubMed 48 Moes M , Le Béchec A , Crespo I , et al. A novel network integrating a miRNA-203/SNAI1 feedback loop which regulates epithelial to mesenchymal transition . PLoS One 2012 ; 7 ( 4 ): e35440 . Google Scholar CrossRef Search ADS PubMed 49 Siemens H , Jackstadt R , Hünten S , et al. miR-34 and SNAIL form a double-negative feedback loop to regulate epithelial-mesenchymal transitions . Cell Cycle 2011 ; 10 : 4256 – 71 . Google Scholar CrossRef Search ADS PubMed 50 Fre S , Pallavi SK , Huyghe M , et al. Notch and Wnt signals cooperatively control cell proliferation and tumorigenesis in the intestine . Proc Natl Acad Sci 2009 ; 106 ( 15 ): 6309 – 14 . Google Scholar CrossRef Search ADS PubMed 51 Bonnet E , Calzone L , Rovera D , et al. BiNoM 2.0, a Cytoscape plugin for accessing and analyzing pathways using standard systems biology formats . BMC Syst Biol 2013 ; 7 : 18 . Google Scholar CrossRef Search ADS PubMed 52 Chanrion M , Kuperstein I , Barrière C , et al. Concomitant Notch activation and p53 deletion trigger epithelial-to-mesenchymal transition and metastasis in mouse gut . Nat Commun 2014 ; 5 : 5005 . Google Scholar CrossRef Search ADS PubMed 53 Kuperstein I , Robine S , Zinovyev A. Network biology elucidates metastatic colon cancer mechanisms . Cell Cycle 2015 ; 14 ( 14 ): 2189 – 90 . Google Scholar CrossRef Search ADS PubMed 54 Costa A , Kieffer Y , Scholer-Dahirel A , et al. Fibroblast heterogeneity and immunosuppressive environment in human breast cancer . Cancer Cell 2018 ; 33 ( 3 ): 463 – 79.e10 . Google Scholar CrossRef Search ADS PubMed 55 Lonjou C , Damiola F , Moissonnier M , et al. Investigation of DNA repair-related SNPs underlying susceptibility to papillary thyroid carcinoma reveals MGMT as a novel candidate gene in Belarusian children exposed to radiation . BMC Cancer 2017 ; 17 : 1 – 9 . Google Scholar CrossRef Search ADS PubMed 56 Fujita KA , Ostaszewski M , Matsuoka Y , et al. Integrating pathways of Parkinson’s disease in a molecular interaction map . Mol Neurobiol 2014 ; 49 ( 1 ): 88 – 102 . Google Scholar CrossRef Search ADS PubMed 57 Matsuoka Y , Matsumae H , Katoh M , et al. A comprehensive map of the influenza A virus replication cycle . BMC Syst Biol 2013 ; 7 : 97. Google Scholar CrossRef Search ADS PubMed 58 Cerami EG , Gross BE , Demir E , et al. Pathway Commons, a web resource for biological pathway data . Nucleic Acids Res 2011 ; 39 : D685 – 90 . Google Scholar CrossRef Search ADS PubMed 59 Slenter DN , Kutmon M , Hanspers K , et al. WikiPathways: a multifaceted pathway database bridging metabolomics to other omics research . Nucleic Acids Res 2018 ; 46 : D661 – 7 . Google Scholar CrossRef Search ADS PubMed 60 Shannon P , Markiel A , Ozier O , et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks . Genome Res 2003 ; 13 : 2498 – 504 . Google Scholar CrossRef Search ADS PubMed 61 Noronha A , Daníelsdóttir AD , Gawron P , et al. ReconMap: an interactive visualization of human metabolism . Bioinformatics 2016 ; 33 : 605 – 7 . 62 Heirendt L , Arreckx S , Pfau T , et al. Creation and analysis of biochemical constraint-based models: the COBRA Toolbox v3.0 . arXiv 2017 ;1710.04038. 63 Ghosh S , Matsuoka Y , Asai Y , et al. Toward an integrated software platform for systems pharmacology . Biopharm Drug Dispos 2013 ; 34 : 508 – 26 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Briefings in BioinformaticsOxford University Press

Published: May 3, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off