iCAVE: an open source tool for immersive 3D visualization of complex biomolecular interaction networks

Visualizations of biomolecular networks assist in systems-level data exploration in myriad cellular processes in health and disease. While these networks are increasingly informed by data generated from high-throughout (HT) experiments, current tools do not adequately scale with concomitant increase in their size and complexity. We present an open-source software platform, interactome-CAVE, (iCAVE), that leverages stereoscopic (3D) immersive display technologies for visualizing complex biomolecular interaction networks. Users can explore networks (i) in 3D in any computer and (ii) in immersive 3D in any computer with an appropriate graphics card as well as in CAVE environments. iCAVE includes new 3D network layout algorithms in addition to extensions of known 2D network layout, clustering and edge-bundling algorithms to the 3D space, to assist in understanding the underlying structures in large, dense, layered or clustered networks. Users can perform simultaneous queries of several databases within iCAVE or visualize their own networks (e.g. disease, drug, protein, metabolite, phenotype, genotype) utilizing directionality, weight or other properties by using different property settings. iCAVE has modular structure to allow rapid development by the addition of algorithms, datasets or features without affecting other parts of the code. Overall, iCAVE is a freely available open source tool to help gain novel insights from complex HT datasets.


Introduction
Interaction networks are one of the primary visual metaphors for communicating and understanding -omics data at a systems level. From cellular organisms to human society, they provide critical clues on systems--level behavior [1][2][3] , and in biomedicine they are essential for understanding both normal 4,5 and disease states [6][7][8][9] , and instrumental for drug discovery [10][11][12] as well as biomarker identification [13][14][15] . Changes in networks have been shown to help in prognosis for breast cancer patients 6 , analyzing systematic inflammation in humans 8 , or studying emerging tumor markers 16 . Network visualizations have thus become key tools in basic and translational biomedical research. Consequently, there is an abundance of tools for their interpretation and exploration 17,18 . Many of these tools are also coupled with public databases, allowing data visualization and interpretation in the context of previous knowledge 17 . In fact, currently there are more than five hundred resources listed at http://pathguide.org that provide access to thousands of networks, cataloging millions of interactions between biomolecules 19 .
Among the currently available biomolecular network visualization tools, the most popular is an open--source and freely available tool, Cytoscape 20 , which enables explorations with different filter, layout, color and cluster options, and includes estimations of network topology parameters and centrality measures. There are also a number of JavaScript network visualization libraries (e.g. sigma.js http://sigmajs.org/), and software packages (e.g. iGraph http://igraph.org) on the web. However, the currently available layout algorithms in these libraries and employed in Cytoscape 21 , in addition to other tools like Ingenuity 22 , Osprey 23 , VisANT 24 , BINA 25 to name a few, are inherently limited by the number of molecules and interactions that can be displayed in the 2D--space of a screen, and the associated layout and representation challenges. As the databases that supplement biomolecular interaction networks are growing at an unprecedented rate due to the increasing size and complexity of --omics experimental techniques, innovations are necessary to address the challenges the concomitant large, dense, and/or multi--layered networks present. Furthermore, because increasingly powerful technologies have enabled the collection of data from multiple types of cellular events simultaneously, in order to achieve better understandings of such complex processes, it may be necessary to maximally integrate data across multiple dimensions, pushing the limits of current visualization tools.
New visualization solutions can bring substantial benefits by improving our understanding of complex mechanisms in human disease, reducing the time to discovery and diagnosis. One approach has been to shift to 3D, and even to immersive 3D space by employing cave automatic VR environments (CAVEs), which are immersive VR environments that include projectors directed to several walls of a room--sized cube 26 . While the relative benefits of immersive 3D in network visualization are still being debated [27][28][29] , a recent study that visualized the same network in an immersive 3D CAVE environment vs. 2D display identified a global network property due to the additional features of the CAVE, and quantitatively validated this result by comparing to 1000 random permutations of networks of the same size and distribution 26 . These beneficial features were stereoscopic visualization, magnification and wide field--of--view 26 . Stereoscopic visualization creates the illusion that objects seen are volumes in 3D--space, which results from the projection of separate left and right eye images of each object, and then combining them in stereo--enabled eyeglasses. Additionally, motion sensors on the eyeglasses enable automatic detection of the user's location if she is moving, and adjust the image perspective, and hand--held controls enable user-network interactions such as zooming and rotating the view. However, the study has not introduced a tool to the community. Also, the researchers only utilized an extension of the standard force-directed network layout of Fruchterman--Reingold 30 to 3D, and did not test performance of additional existing or new layout or clustering algorithms or topology parameter measurements. To perform similar comparative immersive 3D vs. 2D studies, to test alternative algorithms or to analyze networks in immersive 3D other researchers would need to write their own code. They would also need a CAVE, which is a substantial investment to build and maintain. Note that while several network visualization tools incorporate 3D layouts 31-33 , they are not immersive 3D, meaning that they do not have interoperation capability with Virtual Reality (VR) technologies, and their displays are in 2D. For example, Arena 3D 31 mixes 3D properties with 2D, by arranging data in multilayered graphs in 2D, with each layer representing a different data type. While the tool then implements several layout and clustering algorithms for each layer, and layers can be zoomed in/out and rotated, it does not offer global layout and clustering algorithms to make full use of the third dimension: each layer still has a 2D layout on its surface. Directed edges are also not supported 31 . 3DScapeCS 32 is a Cytoscape PlugIn written in Java, with built--in network layout algorithms that are extensions of the classic 2D force--directed layouts. The tool does not allow users to add new layouts or functionalities 30 and does not utilize 3D effects to help improve comprehension, such as transparency, or advanced shadow effects. BioLayoutExpress 33 (current name Miru) is a stand--alone 3D application specifically for gene expression networks, which currently provides three network layout algorithms, a single clustering method, no edge bundling and a limited number of network topology statistics, which cannot be saved by the user and does not allow directional edges. Importantly, the tool has a licensing fee and thus is not freely available. In summary, 3D network visualization field is still somewhat nascent. We need freely available open--source tools for biologists to visualize their biomolecular networks, and at the same time for algorithm developers to add and test their methods that take advantage of the third dimension. This will help the community to understand how best to exploit features unique to 3D in biomolecular network exploration, providing insights to the ongoing debate on the advantages of (immersive) 3D vs. 2D. Here, we introduce interactome--CAVE (iCAVE), an open--source tool for 3D and immersive 3D visualization of complex networks. It is designed primarily to assist biomedical researchers in data exploration, though it can be used in any field that involves networks. iCAVE development is made possible by the continuous evolution of data analysis tools in VR, stereoscopic visualization and emerging 3D technologies. Use of VR technology in life sciences research is still nascent [34][35][36][37] , and so far do not include freely available open source tools for biomolecular network visualizations, mainly due to the limited portability of the technology to personal computers until recently. We designed iCAVE without this limitation, by taking advantage of recent advances in computer graphics hardware, software and content creation that are leading to a proliferation of stereoscopic visualization capabilities in personal computing. Driven primarily by gaming and movie industries, computers can now be upgraded to display high quality stereoscopic 3D visuals with wireless glasses and advanced software 38 at nominal prices. As most scientific computers are becoming 3D-capable and the glasses are going mainstream, iCAVE is on the leading edge of this larger trend in the evolution of visual computing technology. While iCAVE works in CAVE environments, its real benefit comes from enabling immersive 3D network visualization in stereo--enabled computers. Therefore, if a computer is equipped with stereo capabilities, users can display immersive 3D visualizations. If the computer is not equipped with stereo capabilities (or if users choose to turn off stereo), iCAVE provides an interactive (non--immersive) 3D environment that still offers most of its features. As a visualization and analysis tool, iCAVE enables network explorations in hypothesis--driven contexts that is flexible, collaborative and user friendly. It introduces new 3D algorithms that are built--in for laying out nodes and their connections in 3D space (hemispherical and multi--level layouts) as well as graph clustering algorithms for clustering the nodes based on network structure and connectivity and then laying out the resulting clusters in 3D space. It also allows the users to add their own layout or clustering algorithms. Furthermore, users can visually integrate multiple clusters or data types from several databases within the same graph as a multi--layered network (e.g. metabolomic, proteomic, genomic, GWAS--disease, protein--drug interactions). iCAVE reports several network topological properties and centrality measure statistics as 2D reports (now shown). While not extensive, it also includes several built--in databases, to assist in preliminary mapping of High--Throughput (HT) experimental data in the early discovery phase of network building. Customizable color, texture, size and layout options assist in displaying maximum information in a graph in an optimized manner. Edges can be in user--defined colors, weights and directions and can be bundled together for simplified views. Data can be uploaded to iCAVE in a simple tab--delimited text file format; output can be saved as 2D snapshots or movies configured with user--defined rotation, zoom and speed parameters. Results iCAVE users can utilize features that are unique to 3D or immersive 3D visualizations and test whether these improve the quality of their network exploration. These features include stereoscopic visualization, wide field of view, magnification, motion sensors and hand--held controls, as described in introduction section. For example, consider rendering a 2D biomolecular network affected by genomic alterations in glioblastoma 39 . In this example, the network layout algorithm is a simple 3D extension of the classical force--directed Fruchterman--Reingold 30 (Fig 1). Note that instead of the static 2D network in Fig. 1A, iCAVE users experience full 3D depth perception at the comfort of their own stereo--equipped computer (Fig.  1B), or inside a CAVE (Fig.  1D). Furthermore, users without a stereo--equipped computer are also able interact with the 3D network: by using their mouse (in lieu of hand--held controls), they can zoom or rotate the network to a view without occlusions. This ability to rotate and zoom enables viewing of the network from different view angles, such as the screenshot in Fig. 1C. Such easy exploration enables users to visually identify a feature unique to the topology of this example network. The network feature was not intuitive from the original 2D layout in Fig.  1A: nodes CBL and SPRY2 (with *) are connectors between two dense network regions (modules) (Fig 1A--C). A targeted attack to these genes can split the network into two. Such discoveries of network topological features, among others, give a richer, more intuitive and ultimately more insightful understanding of network data.

Addressing Large Networks.
When exploring large networks, researchers may miss important characteristics if they cannot interact with the complete network. In the simplest case, the nodes may form (i) dense sub-networks that are interconnected by a small number of connector nodes which render them critical or (ii) multiple networks (often one giant and few smaller ones) where the smaller sub--networks may represent functional groups of importance, such as a critical enzyme complexes. Thus, there are benefits visualizing the complete network even if it is very large. At the same time, while the human brain has a remarkable capacity to visually identify patterns, enabling interpretation of data, visualizations of large networks may exhibit problems with display clutter, molecular positioning or perceptual tension, which may lead the user to misinterpret closely positioned molecules as related 47 . Such misinterpretations are inherent in the limitations of human visual perception, and have been well--studied in (Gestalt) psychology: people tend to organize visual elements into groups 48 . Using 3D layouts, the elements that appear to form a pattern because of their visual positioning in one viewpoint can be interpreted correctly by rotating the image to a different viewpoint (as shown in Fig. 1). Furthermore, in networks that are denser or larger than that of Fig.1, the potential 2D hairball effect can obscure important interactions. Using iCAVE, the user can simply navigate to a view without occlusions by moving her head, rotating the image, and zooming in or out, so that edge--crossings causing the hairball effect in 2D are eliminated. To further address cluttering problem, iCAVE provides an edge--bundled display 49 option for visually bundling adjacent edges together, analogous to bundling of electrical wires or cables. Bundling is extremely useful in identifying global patterns in very large networks and can suggest vulnerabilities as targets. There are several layout algorithms built--in within iCAVE to address the molecular positioning problem; depending on the topology of the network to be visualized, one layout may work better than another. We suggest testing each layout to see which works best. We provide examples of how these features can help with exploring a network in the following sections. New biological insights gained from networks with known 3D physical coordinates. Users can generate visualizations of physically constrained networks at multiple scales, ranging from proteins ( Fig. 2A) to the whole brain (Fig. 2B). Visualizations that employ physical positions of a 3D network coupled with edge bundling can provide insights during hypothesis generation. For example, Fig. 2A represents a snapshot of bacterial leucine transporter (LeuT) residue correlation network, where the nodes represent 3D coordinates of alpha--carbon of a residue and edges represent top 3,000 (Pearson) correlations between residue pairs from a Molecular Dynamics simulation (from Michael LeVine, personal communication). Remarkably, bundling the edges of this network enables the representation of highest density correlation highways that travel through substrate permeation core in protein center, connecting extracellular and intracellular domains. While these highways are visually fascinating, they also enable users to identify specific residues that have dense correlation highways outside of the protein core, which are unexpected. These residues may have previously unidentified importance in protein structure and are potential candidates for follow--up studies. Automated Layouts utilize 3D for molecular positioning. The topology of cellular and disease networks tend to follow basic and reproducible organizing principles, and navigating the entire network provides a good initial understanding of such a network. The network layout algorithm must address the complex problem of arranging the nodes to clearly disseminate the topology, and at the same time be visually pleasant and user--friendly. iCAVE offers several algorithmic options for network layout to achieve these aims. Due to user familiarity, we extended several variations of the force--directed layout approach to 3D: (i) the classical force--directed algorithm 30 treats the network as a physical system with edges analogous to springs and nodes to electrically charged particles that repel each other. The final layout is established at the state at which the repulsive and attractive forces balance each other 40 (see Figure 3); (ii) Lin--log layout 41 is better suited for larger networks because it keeps highly connected nodes in close proximity with minimal number of edge crossings; and (iii) hybrid force-directed layout 42 partitions the graph into smaller units before applying the force--directed algorithm (see Methods). We further implemented two novel layout algorithms to take full advantage of immersive 3D: Semantic levels layout algorithm segregates the network into separate layers (default 7) in the third dimension. The layout within individual layer is calculated using a 3D extension of the force-directed approach. Semantic layers layout can be especially useful for user--defined networks where the number of layers and node assignments to layers can correspond to different data types (e.g. see a 2D projection in Figure 5 and 3D video in Supplementary Video 3, where layer1: genes; layer2: diseases; layer3: drugs).
Hemispherical layout is a novel layout algorithm we have developed, that positions the network on the surface of a 3D hemisphere. The most connected node is positioned at the top center of the hemisphere. Then, the whole hemisphere surface is populated based on a decreasing rank--order of connectivity. The node positions are fixed and the edges are drawn on the hemisphere surface (e.g. see a 2D projection in Figure 6C and 3D video in Supplementary Video 4). Each layout algorithm has unique strengths and we recommend the user to test different options. Semantic layout is often ideal for hierarchical networks. Force--directed layout often captures the essence of large networks. Hemispherical layout leads to clean images with optional edge bundling ( Fig. 6C and Supplementary Video 4).

Statistics on Network Topological Properties.
Most real--world networks exhibit substantial and non--trivial topological features, where connections are neither purely regular nor random. iCAVE automatically generates and reports network topology statistics and centrality measures both graphically and in tabular form (not shown). These statistics include the number of nodes, the number of edges, network diameter, node--betweenness centrality, closeness centrality, neighborhood connectivity, shortest path, topological coefficient, and node degree distribution properties of the network.

COMBO Database for Simultaneous Query of Multiple Data Types.
The currently publicly available biomolecular interaction data are often contained in databases that are massive in size 19 . While not comprehensive, iCAVE combines data from multiple resources into a single COMBO repository to enable quick queries. This includes protein--protein interaction databases Human Protein Reference Database (http://www.hprd.org) and intAct (http://www.ebi.ac.uk/intact), disease and associated gene variants database (http://www.genome.gov/gwastudies); and drug--target databases STITCH (http://stich.embl.de) and DRUGBANK (http://www.drugbank.ca). Pathways database SuperPathway is stored separately (personal communication with Josh Stuart, UCSD). Users can add their own databases without affecting other parts of the code. Details on the COMBO database are given in Supplementary Table 1.

Visualizing Multiple Layers of Information.
Effective usage of genomic information can depend on finding systems--level connections between multiple types of information, such as that of between genomic variation, disease and drugs [43][44][45][46] . Visualizing such data by using semantic layout can assist in exploration in higher--level organization, all in one graph. User can pick a gene (e.g. AHR, dark blue, Fig  5A), query the COMBO database for diseases associated with its variants (purple); identify drugs that target it (green) and drug candidates that may target (light blue) due to guilt by association for having common targets with AHR--targeting drugs. These serve as initial candidates for subsequent binding site characterization. Querying COMBO database further generates a hierarchical network of proteins that interact with AHR ( Figure  5B, middle layer), diseases associated with gene variants of AHR--interacting proteins (purple) and AHR targeting drugs (green). Illustrative Examples. Example 1. Visualizing the complete global network, even if it is very large, can enable visual identification of a pattern. For example, consider a large probabilistic causal network constructed from human omental adipose tissue in a morbidly obese patient cohort in Fig. 4A. The network consists of 7,601 nodes, 13,979 edges 50 . Nodes are the genes expressed in tissue; edges are derived from a Bayesian network reconstruction algorithm that leverages DNA variation for causality. Here, we highlight nodes that represent a signature of genes causally associated with inflammatory bowel disease (IBD) SNPs or disease pathways. Notice that within this global view of the massive network, there is a pattern of the IBD genes clustering together, which visually supports the hypothesis of functional relatedness. Example 2. While force--directed layout algorithms can help identify global patterns as in previous example, if the interaction network has a hierarchy, the semantic layers layout can help visualize the hierarchical nature of the interactions easily. For example, Fig 4B displays the global view of network generated from The Encyclopedia of DNA Elements (ENCODE https://www.encodeproject.org/) study data. The ENCODE Consortium is generating a comprehensive parts list of the human genome functional elements, including those that control active genes, such as transcription factors (TFs). Utilizing these unprecedented volumes of data, Gerstein and co--workers have generated the massive network in Fig  4B  that  includes  119 TFs that target 9,057 genes (nodes) via 26,037 interactions (edges) 51 . Using force--directed layouts, users can capture the general network structure and differentiate a TF from its neighbors by zooming in/out, adding labels to that specific TF, etc, as well as obtain statistics on its network centrality and other global topological properties as they pertain to the network. However, the semantic layers layout is useful in visualizing the hierarchical nature of this network, integrating TF, non--coding RNA (ncRNA), miRNA and protein--protein interaction data (Fig 4D and Supplementary Video 2). In this figure, network connectivity and hierarchy reflects genomic properties:top level TF--binding correlates with target expression, mid--level contains 'information flow bottlenecks' and connections with miRNA and distal regions, revealing ideal drug targets. Such multi--layered heterogenous information integration assists in differentiating intra--level interconnections as well as inter--level edge types and node labels. Note that nodes in each layer are also arranged in 3D using 3D force--directed layout. Example 3. Visualizing the global network of interactions while scaling or coloring a subset of the nodes based on their specific properties can enable hypothesis support. In this example, the visualization helps support the principle that functionally significant and highly conserved genes tend to be more central in physical protein--protein and regulatory networks 52 . Based on this hypothesis, Fig. 4C visualizes a network of tolerance to loss--of--function (LoF) mutations and evolutionary conservation, with nodes for (LoF) tolerant (blue) and essential genes (red) easily distinguishable 52 . Node size is based on the degree centrality of a gene 7 . While essential genes tend to be bigger and central, LoF--tolerant genes are smaller and located in the periphery. Both the 2D snapshot ( Figure 4C) and 3D Supplementary Video 1 provide clear visualizations of this complex data that lead to easy interpretation. Note that we have published an iCAVE--generated visualization of a network with similar properties that enabled help support this hypothesis 53 .

Graph Clustering To Identify Network Motifs.
Clustering is critically important during network exploration, as biomolecules that cluster together tend be functionally related. iCAVE offers the following graph clustering algorithms: Edge--Betweenness clustering (EBC). The number of shortest paths going through a particular edge is EB. An edge with a high EB value connects multiple communities. At each step, the EBC algorithm removes the edge with the highest EB value until it has optimized a modularity metric on how unlikely the in--cluster degree of a node is in comparison to a random edge. EBC 54 is an attractive algorithm since it does not require an estimate of the number of clusters a priori, unlike a majority of existing graph clustering algorithms. Markov clustering (MCL) 55 is a scalable and unsupervised algorithm which assumes that the number of intra--cluster connections is large and inter--cluster connections is small. It is based on a Network diameter is the length of shortest path between two farthest nodes. Unconnected nodes are not considered. Irregular networks usually have small diameters, while regular networks have large diameters. Betweenness centrality is a global metric on the importance of a node, which is equal to the number of shortest paths from all vertices to all others that pass through that node, calculating the load on a node 67 . Real world scale--free networks usually involve short path lengths across the network, and a few nodes have high betweenness--centrality. Connector or high--traffic biomolecules that are vulnerable to targeted attacks, usually suggest potential non--hub drug targets 68,69,70 . Shared nearest neighbors: A similarity metric based on the sharing of nearest neighbors between any two nodes. Particularly useful in network topology--based motif, sub--graph or cluster identification. Shortest paths: Quantifies the importance of a node within the network, calculated by the number of shortest paths going through the node. Purely random graphs exhibit a small average shortest path length (~ the logarithm of the number of nodes) along with a small clustering coefficient.

Layout Algorithms.
A graph G(V ={1, . . . , n},E) represents a binary relation E over node set V. iCAVE both extends classical layouts to 3D and offers novel algorithms. Based on the underlying topology, a user can choose the best layout that helps with data interpretation. 1.Force--based layout. The forces acting on each node in classical Fruchterman--Rheingold (FR) algorithm 30 are: where fa(ij)and fr(ij) are attractive and repulsive forces, dij is the distance between nodes i and j, and k is a constant corresponding to the equilibrium edge length. 2. Lin--log layouts. We used r--PloyLog 41 energy model to implement the node--repulsion and edge-repulsion LinLog models. For all r R with r > 0, the node--repulsion energy of a layout p is: where p(u) is the position of node u. Edge--repulsion energy is: where deg(u) is the number of edges incident to node u. At r=3, the 3--PolyLog reduces to FR and at r=1 to LinLog model. LinLog models group nodes according to cut density and the normalized cut, therefore the layout leads to graph clustering.
Hemispherical layout. We place n nodes of a graph G(V= {1,..,n}, E) equally spaced on a single 3D hemisphere surface, reducing the problem to finding a hemispherical node ordering. Coordinates for a node i є V are (xi, yi, zi) є R, at fixed hemisphere radius R: ≤ # Nodes are sorted and placed based on their degree, with the highest degree node at the hemisphere surface center. Algorithm inputs are the number of nodes, the graph center position and hemisphere radius. Hemisphere radius, node sizes, colors and textures are adjustable. 4.Semantic levels layout is ideal for integrative analysis of multiple data resources (e.g. genotype, phenotype, drugs, proteins, metabolites). Initially, FR algorithm is performed in 2D. Then, multiple equidistant levels (default =7) are created in the z--dimension. Based on network topology, we consecutively assign the nodes to one of the layers. iCAVE user--interface allows the manual manipulation of the number of layers and the distance between them. If layers are not predefined, we suggest experimenting with different options. 42 Original version of this algorithm is extremely computationally intensive, so we implemented a simplified version, reducing the run time at the expense of visualization quality. Our version has three steps: (i) position nodes randomly; (ii) partition the resulting graph; (iii) apply FR 30 algorithm separately on each subgraph. The partitioning step splits the graph into two sub--graphs (A and B) of equal sizes. This requires minimizing the cut size, by calculating the second Eigenvector (Fiedler vector) λ of the following:

5.Hybrid force directed layout
is the Laplacian of graph G. The power--iteration algorithm solves for λ. Edge--Betweenness (EB) Clustering Algorithm: An edge with a high EB value potentially connects two or more communities. The edge with the highest EB value is removed at each step. The number of edges to be removed is user--defined ( with a default of 0.2 times the number of edges). Any edge that leads to a single--node cluster is not removed. Edge bundling algorithm is based on application of forces (electrostatic and spring) on an edge subdivided into multiple points. Edge compatibility metrics edge angle, scale (length), position and visual compatibility are multiplied for total compatibility. If two edges are compatible above a threshold, forces are calculated and added to each subdivision, and those subdivisions are bundled together.
Medical College of Cornell University (to V.L. and Z.H.G). We wish to thank Dr. Jian Sun and Jason Banfelder for programming assistance and constructive comments. To ensure that the proposed visualization platform meets the needs of translational and basic science researchers, we relied heavily on the advice, feedback and discussions from representative end--users including Mark Gerstein, Alex Lash, Michael LeVine, Chris Sander, Eric Schadt and Karsten Suhre.   User experiencing a full 3D depth perception of the same network with iCAVE while using stereoscopic glasses on his desktop. iCAVE display is generated with force--directed layout algorithm. Notions of edge crossings that create a hairball effect in 2D have little meaning in 3D, as the user can navigate to a view without occlusions, solving the visual clutter problem. c. A screenshot from the 3D display generated with the force--directed layout. This network is generated without a priori knowledge of the network topology; however, it readily identifies hubs, connectors and modules, such as the connectors between two dense regions of the network (highlighted with a (*) in both panels a and c). d. Immersive visualization in a CAVE environment, with one user inside the data space. While the photos only capture images reflected on the interaction walls of the CAVE, the user is interacting with a virtual 3D image. In both b and d, the options to zoom and rotate the network with a mouse (or wand) click helps the user focus on a particular hub or module of interest within the network. While the addition of the third dimension gives a richer, more intuitive and ultimately more meaningful understanding of the network--represented data, the 3D layout brings a completely new modality into network visualizations, with clean, easy to use and understandable layouts. Nodes represent 3D coordinates of alpha--carbon of a residue; edges represent top 3,000 (Pearson) correlations between residue pairs, where the input is 3D coordinates & correlation scores. Surprisingly, 3D visualization with edge bundling enables representation of highest density correlations (correlation highways) that travel through the substrate permeation pore in protein center, connecting extracellular and intracellular domains. Correlation highways at the pore are visually fascinating and biophysically intuitive. Information highways of some residues outside the pore reveal unexpected structural importance (data courtesy of Weinstein Lab, Weill--Cornell). b. Living Human Brain Connectivity. iCAVE visualization of brain regions as nodes, labeled by anatomical region name. Edges show connectivity, and bundling shows connectivity highways. Datasets from Diffusion Tensor Imaging of left hemisphere scanned ith Siemens 1.5Tesla and generated by Fiber Assignment by Continuous Tracking tractography using U. of California, Los Angeles (UCLA) Multimodal Connectivity Package connectivity matrix module. Database is powered by the Human Connectome Project, which aims to provide an unparalleled compilation of neural data to achieve never before realized conclusions on the living human brain. Figure 3: a. A directed and weighted signaling network using semantic layout. Edge color represents activation (red) or repression (green). Color alteration frequency represents weight