Tools for visualization and analysis of molecular networks, pathways, and -omics data.

Biological pathways have become the standard way to represent the coordinated reactions and actions of a series of molecules in a cell. A series of interconnected pathways is referred to as a biological network, which denotes a more holistic view on the entanglement of cellular reactions. Biological pathways and networks are not only an appropriate approach to visualize molecular reactions. They have also become one leading method in -omics data analysis and visualization. Here, we review a set of pathway and network visualization and analysis methods and take a look at potential future developments in the field.

Introduction

The study of biological pathways is a key to understand the different processes inside a cell: proteins exert their function not in isolation but in a tightly controlled network of interactions and reactions. Activation of a pathway typically leads to a change of state in the cell: as examples, a set of genes is transcribed, new molecules are assembled, molecules are secreted, or new signals are recognized. Pathways come in different flavors, depending on their functions in the cell – the three main types are metabolic pathways, gene regulatory pathways, and signaling pathways.

Biological pathways are generally derived from experimental work in cell culture or model organisms. Many different studies are collected and integrated over time by initiatives like KEGG,1 Reactome,2 PANTHER pathway,3 BioCyc,4 or the user-curated WikiPathways5 (Table 1). The KEGG pathway database was first published in the year 1997 and has since been updated regularly. It is to our knowledge the largest pathway resource, with metabolic and signaling pathways from a pleiotropy of organisms. Reactome has pathways for a selected number of organisms, representing mostly the commonly used model organisms. BioCyc has a large collections of metabolic and regulatory pathways for ~5,500 organisms, with components such as EcoCyc and MetaCyc,6 or HumanCyc. PANTHER pathway is a rather small resource, with currently 176 primarily signaling pathways. PANTHER pathway, however, encourages user-curation and user-creation of novel pathways. WikiPathways works under the same principles: users can contribute to the open-source pathway resource by suggesting new pathways to be added or by creating novel pathways. WikiPathways offers a good selection of organisms. Its largest organism-specific pathway resource is the one for Homo sapiens, with currently 633 available pathways.

Table 1

Pathway resources

Name	URL	Formats
KEGG1	http://www.genome.jp/kegg/	BioPAX, png, KGML
Reactome2	http://www.reactome.org/	BioPAX, png, pdf
Pathway Commons7	http://www.pathwaycommons.org/	BioPAX, Sif, png
PANTHER pathway3	http://www.pantherdb.org/pathway/	BioPAX, SBML
WikiPathways5	http://www.wikipathways.org/	BioPAX, svg, png, pdf, gpml
Nature/NCI PathwayInteractionDatabase63	http://pid.nci.nih.gov/	BioPAX, jpg, svg
BioCyc4	http://biocyc.org/	BioPAX, png, SBML
INOH84	http://inoh.hgc.jp/	BioPAX, INOH (xml)
Netpath85	http://www.netpath.org/	BioPAX, SBML, PSI-MI
PharmGKB86	http://www.pharmgkb.org/	BioPAX, pdf, gpml

Abbreviations: BioPAX, Biological Pathway Exchange; KGML, KEGG Markup Language; PSI-MI, Proteomics Standards Initiative Molecular Interaction; SBML, Systems Biology Markup Language; NCI, National Cancer Institute; INOH, Integrating Network Objects with Hierarchies; PharmGKB, Pharmacogenomics Knowledge Base; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Defining novel pathways is however a very labor- and time-intensive task. Furthermore, we cannot assume that currently existing pathway definitions are complete: over time, it has rather become clear that pathways are much more complicated than initially believed. They form a tightly regulated network of reactions, with many cross-reactions between our simpler definitions of individual biological pathways.

A possible solution to circumvent incomplete pathway definitions is the usage of curated protein–protein interaction networks. Biological networks typically contain data from protein interaction studies in addition to curated biological pathways. They are believed to represent a more complete view of the complex, biological network within a cell. Integrated biological network resources include, for instance, Pathway Commons,7 ConsensusPathDB,8 or similar meta-databases. However, interaction networks do typically not provide information on known biological pathways. Therefore, either the network has to be enriched for biological pathways or its structure has to be analyzed for more tightly connected node clusters.

Pathways are a fundamental part of interpreting -omics data, as they provide the biological context for a given observation. Hardly any -omics study is published nowadays without integrating the data with pathway or network information. A general application of biological pathways or networks in -omics data analysis is to integrate them with numerical data from an -omics screen. To select relevant pathways for visualization, users commonly perform pathway enrichment analysis on their hit lists.

The integrative analysis of -omics data, however, increases the complexity, since it adds different dimensions to the data. Moreover, integrating -omics data boosts data size. With growing complexity emerges the problem on how to make the data digestible for the human mind. Visual reprocessing is still one of the most forward ways to comprehend highly complex data, and thus the importance of scientific visualization, since it enhances our understanding of biological data.

A vast variety of different pathway and network visualization and analysis tools exist. In this review, we introduce a few concepts and tools that are in our view useful for analyzing and visualizing pathway and molecular network data. We further discuss general problems and future directions of biological data visualization.

Pathway-based visualization and analysis

In principle, two categories of pathway visualizations exist: first, those that allow a simple and static representation of pathways (such as static images or pdf files), which many pathway resources already offer. Though practical for providing pathway information “at a glance”, their use is limited for -omics data analysis.

The second type of pathway visualization allows users to integrate different data types with pathways by rendering user data on top of pathway maps (Table 2). The KEGG pathway web interface (http://www.genome.jp/kegg/tool/map_pathway2.html),9 provides the possibility to color code individual nodes within a pathway. The TrED database,10 for instance, has implemented the KEGG Mapper in its database to visualize metabolic pathways.

Table 2

Pathway and network visualization and analysis software

Name	Category	Supported pathway resources
KEGG Mapper9	Web	KEGG
KEGG-based Pathway Visualization Tool for Complex Omics Data11	Web	KEGG
MEGU12	Web	KEGG
Pathway Projector13	Web	KEGG
KEGGViewer15	Web, BioJS	KEGG
PathVisio17	Desktop app	WikiPathways, Reactome
Reactome2	Web	Reactome
Pathview19	R-package	KEGG
iPath22/iPath223	Web	KEGG
WikiPathways App44	Cytoscape plugin	WikiPathways
KGMLreader/KEGGscape46	Cytoscape plugin	KEGG
CyKEGGParser45	Cytoscape plugin	KEGG
Reactome FI62	Cytoscape plugin	Reactome
CluePedia48	Cytoscape plugin	KEGG, Reactome
The SEED31	Web	KEGG
CytoSEED33	Cytoscape plugin
COBRA Toolbox34	Matlab toolbox	Reads SBML-
COBRApy35	Python toolbox	formatted models

Abbreviations: KEGG, Kyoto Encyclopedia of Genes and Genomes; BioJS, BioJavaScript; Reactome FI, Reactome Functional Interaction.

The KEGG-based Pathway Visualization Tool for Complex Omics Data11 is another practical, user-friendly method to visualize expression data on KEGG pathways. Its functionality is more complex than the pathway viewer offered by KEGG. Proteins, as well as substrates, can be shaded according to a numerical value for more than one condition. The reader might turn to the website of this visualization tool, http://www.g-language.org/data/mar-ray/,11 where a comprehensive analysis of expression data in Escherichia coli using the software has been performed and which gives a very good impression on the software’s potential.

MEGU is a web-based visualization application that allows users to map multiple layers of -omics data simultaneously.12 MEGU, in addition, provides integrated pathways by combination of several individual pathways. The MEGU website also demonstrates its usability on a comprehensive analysis of E. coli expression data. Visualization is, in this case, more plastic; however it is also more crowded, as several gene names are sometimes provided per node.

BioCyc4 offers a collection of pathways in a variety of organisms for analyzing and visualizing metabolic pathways. It allows the user to overlay experimental data on a pathway map by uploading a file or importing datasets from different databases. BioCyc offers in general a comprehensive tool set for genome analysis and comparison, which goes beyond most of the other pathway resources.

Pathway Projector (http://ws.g-language.org/g4/),13 another web-based visualization software, allows users to explore the network of biological pathways in a cell: the user can zoom into the network, render pathways with experimental data, or predict the enriched pathways for a set of user-provided genes. Pathway Projector was, for instance, used to visualize metabolic changes upon nitrogen deficiency of the oil-rich algae Pseudochoricystis ellipsoidea.14

Finally, with KEGGViewer,15 users can integrate expression data with KEGG pathways and create a web-based visualization using the open-source JavaScript library BioJS.16 KEGGViewer is purely web based. Visualization of differential expression can be done for a series of experiments or time points, as the tool enables users to loop over several uploaded conditions. It is so far the only web-based viewer that allows an animation of differential expression events overlaid on pathways.

Further examples include the stand-alone application PathVisio17 and the web front end of Reactome (http://www.reactome.org/PathwayBrowser/#TOOL=AT),2 where the user can upload expression data and color a target pathway accordingly. Furthermore, PathVisio offers the possibility to fully customize the looks of a given pathway. PathVisio was, for instance, applied by Fijten et al18 to construct a customized, liver-specific ligand-activated nuclear receptor pathway. The R-package Pathview19 finally has a broader approach and creates pathway visualizations from additional data types like genomic variation, literature record, and metabolite level. Pathview has, for instance, been applied by Arthur et al,20 for pathway mapping or by Gorvel et al,21 to render pathways using experimental data. Pathview depends on KEGG for pathway information. PathVisio works with WikiPathways, and for human, also with Reactome data.

Besides permitting users to make queries based on formatted text files, the web tool iPath22 and iPath223 allows users to define visual properties like color or transparency for various nodes, similar to PathVisio. iPath/iPath2 has been extensively used for customized pathway rendering and visualization of experimental data and can conveniently be employed to visualize differential usage of pathways in a global pathway map (for typical applications of iPath2, see, for instance, Navid and Almaas24 and Erickson et al25). Wisecaver et al26 used iPath2 for comparative pathway analysis, and Creek et al27 used the software to model metabolic data obtained by liquid chromatography–mass spectrometry. Dutilh et al28 used iPath2 for interpreting microbial Genome-Wide Association Studies (GWAS) studies, and Guzman et al29 employed it to map microRNA sequencing data on metabolic maps.

To demonstrate the usability of some of the above-mentioned tools, we selected four software packages that can render expression data on pathways from KEGG, WikiPathways, or Reactome (Figure 1). We used RNA-seq data of colon cancer patients and compared expression values from control and paracancerous and cancer tissue.30 The resulting log 2 fold changes were overlaid on the Glycolysis pathway.

Figure 1

Visualization of differential expression data on the Glycolysis pathway using different software packages.

Notes: (A) KEGG Mapper and KEGGViewer were used to render -omics data on the KEGG Glycolysis pathway. Upregulated genes are shown in yellow, and downregulated genes are highlighted in blue. (B) PathVisio was used to visualize -omics data on the Glycolysis pathway from WikiPathways. Upregulated genes are colored in red shades, and downregulated genes are shown in green shading. PathVisio is able to combine two datasets on the same pathway map. (C) The Reactome analysis toolbox was used to visualize -omics data on the Glycolysis pathway. Upregulated genes are highlighted in yellow, and downregulated genes are highlighted in blue.

In Figure 1A, we used KEGG Mapper and KEGGViewer as web-based visualization tools for KEGG pathways. KEGG Mapper creates a static image, whereas KEGGViewer allows users to loop over different conditions, making it potentially easier to spot differential events between several conditions or a time series. KEGG Mapper on the other hand visualizes the directionality of reactions, which are omitted in KEGGViewer. For metabolic pathways, the labels in KEGG Mapper are by default EC numbers, while KEGGViewer provides gene names. However, the user can interactively use the pathway map in KEGG Mapper to retrieve detailed information on an individual gene by clicking on its node.

The WikiPathways version of the Glycolysis pathway was created using PathVisio (Figure 1B). PathVisio is a very powerful method to visualize differential events in several conditions on the same map. It keeps directionality of reactions and is able to combine more than one pathway, whereby the edges (= reactions) from different pathways are color coded.

Finally, we used the Reactome analysis tool to visualize differential expression of our gene set on the Glycolysis pathway (Figure 1C). Reactome gives quite a different picture of the pathway, embedding it within its neighbors. This extended visualization can be very useful in providing a more complete picture of differential expression events. It, however, can also be distracting, if the pathway map gets too large, and therefore, too crowded.

The handling of KEGG Mapper, as well as Reactome, was very easy and straightforward. The input format is simple, and KEGG Mapper accepts several identifiers; Reactome even does identifier conversion on the fly. Reac-tome moreover offers a variety of additional information on genes, such as the expression of a gene in different tissues. Both tools are easy to use for beginners and thus represent a good starting point for rendering expression data on pathways. PathVisio and KEGGViewer require more sophisticated handling. PathVisio is a desktop application, and KEGGViewer requires some knowledge of JSON (JavaScript Object Notation) and JavaScript. While the learning curves for the latter programs are steeper, both tools offer useful functionalities, such as looping over conditions (KEGGViewer) or mapping several conditions on the same pathway map (PathVisio).

In general, each pathway visualization tool has advantages and disadvantages. While we give examples only for a very small collection of available pathway mapping tools, we recommend to try out several methods and to regularly look for novel developments in the field of pathway visualization.

It is noteworthy to mention that many software packages for in silico metabolic modeling also allow users to visualize and analyze molecular pathways and networks with -omics data. These include, for instance, SEED (http://seed-viewer.theseed.org/),31 and the COBRA project.32 SEED has a web-based user interface and allows to read in SEED models in Cytoscape using the plugin CytoSEED.33 COBRA can be used via Matlab34 or Python.35 We predict again a steep learning curve, as the tools are tailored for metabolic modeling applications. Depending on the complexity of the research question, however, such tools might offer more functionality.

Molecular networks analysis

Other than pathways, which contain information on directionality of reactions, as well as on small compounds, protein interaction networks are a collection of nodes (proteins) connected to each other (via edges). Biological interaction networks are typically created for one organism and contain information on all known biological interactions within the cell, whereby genetic and physical interactions are collected separately. Biological networks, though being less detailed in information than pathways, are nevertheless useful for analyzing and visualizing -omics data.

The most popular tools for network analysis are Cytoscape36 and Gephi.37 Both are open source and aid network exploration, manipulation, and analysis by providing filtering, layout, coloring, and clustering capabilities.

Cytoscape’s network analysis package38 and Gephi facilitate network analysis by calculating complex network parameters like average clustering coefficients, shortest paths, and node degrees, as well as centrality measures like stress centrality,39,40 betweenness centrality,39 and closeness centrality.41 The calculated parameters can be visualized as histograms and mapped as visualization attributes on the network in the form of node size and color.

Cytoscape and Gephi are able to handle diverse datasets, which can be biological or not. However, Cytoscape tends to be biased toward biological data analysis. It is able to fetch data from I) molecular interaction databases, II) Biomart, and III) different ontologies out of the box. Additionally, Cytoscape is extendible via community developed plugins42 that range from Gene Ontology enrichment43 to clustering to pathway visualization and analysis. For instance, the Cytoscape plugins WikiPathways,44 and CyKEGGParser45 parse and visualize WikiPathways and KEGG pathways, respectively. KGMLReader and KEGGScape46 are plugins provided by KEGG for pathway import and visualization in Cytoscape versions 2 and 3, respectively. Similarly, the Reactome FI (Functional Interaction) plugin47 finds pathways and network patterns related to cancer and other diseases using Reactome data. Its handling is very easy and comparable to the web-based visualization application of Reactome. CluePedia48 is a search tool for new markers potentially linked to known pathways. It is one of the tools that offer the potential to extend our existing pathway definitions.

Gephi, on the other hand, tends to focus on network visualization. It is based on the principle What You See Is What You Get. It is therefore able to generate powerful visualizations. Like Cytoscape, Gephi can be extended via plugins.

To illustrate the visualization capabilities of both, Cytoscape and Gephi, we collected molecular interactions from VirHostNet 2.0,49 a knowledge base dedicated to network-based exploration of molecular virus–host interactions. We created a network of 7,318 nodes and 296,659 edges composed of human and influenza A virus proteins. We further created a subnetwork with influenza A interactors and their first neighbors, which resulted in a network of 3,362 nodes and 19,352 edges. We applied several force-directed layout algorithms to those networks (Figure 2). We chose force-directed layouts as they rely on the structure of the network and thus do not require domain-specific information. These layout types furthermore are visually more appealing as they reduce edge crossing and reveal symmetries within the graph. To visualize a large network is generally problematic. We found the Fruchterman Reingold layout as the most appealing, as it provides some form of structure for very large networks. All chosen layout options are separating the small network in several, more tightly connected clusters, with the Spring layout performing best with respect to visual separation of subclusters.

Figure 2

Network visualization using Gephi and Cytoscape using different layout algorithms.

Notes: The entire human and influenza A interactomes are shown, as well as the first neighbors of the influenza A proteins. Influenza A proteins are colored in orange, and human nodes are shown in green. The size of the nodes indicates the connectivity of the individual proteins.

While Cytoscape and Gephi are popular tools, there is a broad spectrum of alternatives available. For example, sigma. js50 and Cytoscape.js51 are JavaScript libraries for network visualization on the web. Additionally, software packages such as iGraph52 and NetworkX53 further support network analysis and visualization.

Pathway standardization

Pathway resources are annotated and curated separately. It is due to this fact that heterogeneous data models, formats, and application programming interfaces (APIs) are used. This makes the development of pathway visualization and analysis tools as well as the aggregation of pathway data an arduous task. Fortunately, the community using the pathway resources came together and launched the Biological Pathway Exchange (BioPAX) project54 with the aim to define a standard for sharing pathway information.

BioPAX supports I) metabolic and signaling pathways, II) molecular and genetic interactions, and III) gene regulation data. It encapsulates the semantics related to pathways by using controlled vocabularies such as the Gene Ontology55 and the Proteomics Standards Initiative Molecular Interaction56 vocabularies, as well as its own community-defined constructs.

Several tools have been developed to read and process BioPAX files. These include libraries like paxtools,57 rBiopaxParser,58 and BioPAX-pattern,59 and visualization and analysis tools like Cytoscape and cyPath2,60 among others.

However useful, the primary goal of BioPAX is not to satisfy data visualization but to coordinate its forces with the Systems Biology Graphical Notation (SBGN, http://www.sbgn.org/),61 community and to ensure that BioPAX can be visualized through SBGN.

SBGN is a community-developed, standard graphical language for unambiguously representing diverse biochemical and cellular events. One of the key aspects of SBGN is to minimize the effort needed to understand a pathway. It does so by limiting the number of symbols and reusing them when possible. Data providers such as Reactome and PANTHER pathway support SBGN. Additional tools supporting SBGN are listed in http://www.sbgn.org/SBGN_Software.

However, the current standards are still maturing and so not fully adopted. For example, KEGG provides BioPAX- formatted pathways through their FTP server only. Additionally, KEGG does not follow the SBGN specification but rather uses its own graphical representation format KEGG Markup Language (http://www.kegg.jp/kegg/xml/).

Application programming interfaces

An API is a particular set of guidelines and specifications that software applications can follow to communicate with each other. It provides an interface between different applications facilitating their interaction. The combination of different APIs allows developers to create new applications, visualizations, and services9,44,46,62 with use cases beyond the scope of a given data type or data source.

Most pathway data sources1–3,5,7,44,63 provide data access methods such as I) return relevant information about a given pathway (list of participants, list of complexes, etc), II) search for a keyword or pathway in a given database, III) convert identifiers from one format to another, and IV) fetch pathway maps as plain bitmap images or, in some cases, as XML files.

Additionally, a few APIs provide methods for data analysis. For example, through the PANTHER pathway API, it is possible to run an overrepresentation test: by comparing a list of genes to a reference list, the statistical over- or underrepresentation of selected categories is determined (eg, function, process, cellular location, protein class, or pathway). Similarly, Reactome recently developed an API devoted solely to data analysis.

With regard to the retrieval of molecular interaction data, the Proteomics Standard Initiative Common QUery InterfaCe64 provides a standardized web service to access multiple interaction resources such as IntAct65 and STRING.66

Discussion

The amount of data produced by the different -omics fields is rapidly growing, mainly due to the constant development of high-throughput methods. Given the fast rate at which vast amounts of data are produced, the need for new data visualization and analysis methods is undisputed. There are, however, a series of challenges that need to be overcome.

Standard formats allow biological data to be easily exchanged, manipulated, aggregated, and analyzed. Different communities, like the Proteomics Standard Initiative67 and BioPAX, are addressing that need, for proteomics and biological pathways, respectively. However, data integration is still challenging. For instance, mapping between different identifiers such as protein, gene, transcript, or clone is perhaps the most common operation a bioinformatician has to perform. Though there are several tools for mapping identifiers,68–70 significant manual effort is still required. Further development of tools and models able to link and aggregate datasets from various sources and types is crucial to enable detailed analysis and rich visualizations.

Traditionally, networks or graphs are visualized in node-link diagrams, where nodes represent entities and edges represent relationships. Such a representation is intuitive and works well for small networks, but as soon as the network size increases, so does the visualization complexity. Different techniques have been developed to try and enhance such visualizations: node filtering, edge bundling,71,72 different layout algorithms,72–75 and edge lenses are currently available in the toolbox. However, the traditional node-link visualization is limited when applied to large networks.

Alternative network visualizations such as matrix diagrams74,75 and 3D models76–79 are exiting steps towards a hairball-free visualization and may lead to much more interesting visualization techniques than the traditional node-link diagrams.

Matrix-based visualization

Matrix diagrams80,81 are based on the adjacency matrix of a given graph. The main advantage of this kind of visualization is that line crossings are impossible, which leads to a clear visualization. However, the effectiveness of a matrix diagram is heavily dependent on the order of rows and columns. Therefore, patterns may be hidden due to a nonoptimal clustering or ordering algorithm.

3D visualization

Adding an additional dimension enhances the understanding of the topology of a large network. It increases the possibilities of layout elements avoiding undesirable visualization objects, like line crossings. 3D visualizations have been used to analyze biological networks72,73 (aided by BioLayout Express3D) as well as biological pathways.78,79

Virtual reality

Virtual reality offers the possibility of fully exploring complex datasets by using powerful interactive visualizations. For example, Redgraph82 allows users to visualize semantic social networks in a virtual reality-based environment supported by the Computer-Assisted Virtual Environment83 (CAVE). Users can interact with the visualization in order to contextualize graph nodes by clicking on them in order to fetch relevant data.

Other interesting CAVE visualizations permit users to see the world at the molecular level. For example, the gold particle visualization explores the self-assembly of a ligated gold nanoparticle and proteins inside an ionic solution, while another initiative explores a glass fissure computed in nanoscale simulation. Similar approaches could be directed toward the exploration and understanding of diverse biological datasets, such as pathways, gene regulatory networks, and molecular interactions.