See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/262071553

MuxViz: A Tool for Multilayer Analysis and Visualization of Networks

Article · May 2014

DOI: 10.1093/comnet/cnu038 · Source: arXiv

CITATIONS READS

137 938

3 authors, including:

Manlio De Domenico Alex Arenas

Universitat Rovira i Virgili Universitat Rovira i Virgili
123 PUBLICATIONS   5,122 CITATIONS    256 PUBLICATIONS   17,305 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Complex systems effects triggered by destruction in the evolution of altruism. View project

All content following this page was uploaded by Alex Arenas on 30 April 2015.

The user has requested enhancement of the downloaded file.

 Multilayer Analysis and Visualization of Networks

M. De Domenico,1 Mason A. Porter,2 and A. Arenas1

1
Departament d’Enginyeria Informática i Matemátiques,
Universitat Rovira I Virgili, 43007 Tarragona, Spain
2
Oxford Centre for Industrial and Applied Mathematics,
Mathematical Institute, University of Oxford, Oxford,
OX2 6GG; CABDyN Complexity Centre, University of Oxford, Oxford OX1 1HP, UK
Multilayer relationships among and information about biological entities must be accompanied by
the means to analyze, visualize, and obtain insights from such data. We report a methodology and
a collection of algorithms for the analysis of multilayer networks in our new open-source software
(muxViz ). We demonstrate the ability of muxViz to analyze and interactively visualize multilayer
data using empirical genetic and neuronal networks.

In the last two decades, scientific understanding of life Although the aforementioned overlap is an indication
and disease has benefited significantly from the modeling of correlation between a pair of networks, the analysis
arXiv:1405.0843v1 [physics.soc-ph] 5 May 2014

of biological interactions in terms of networks [1–6]. Con- of multilayer biological data would benefit greatly from
nections among genes, proteins, neurons, and other bio- techniques and diagnostics that are able to exploit multi-
logical entities can indicate that they are part of the same plexity (i.e., multiple different ways to interact) in avail-
biological pathway or exhibit similar biological function. able information. Recently, a novel mathematical frame-
The formalism of networks focuses on connectivity and work to model and analyze multilayer relationships and
has now become a paradigmatic way to investigate the their dynamics was developed [14, 15]. One represents
organization and functionality of cells, synaptic connec- the underlying network topology and interaction weights
tivity, and more [7–11]. as a multilayer network, in which entities can exhibit dif-
In parallel, a large variety of computational techniques ferent relationships simultaneously and can exist on dif-
have been developed to analyze (and visualize) networks ferent “layers”. Multilayer networks can encode much
and the biological information that they encode. Such richer information than what is possible using the indi-
methods have become important tools for attempts to vidual layers separately (which is what is usually done).
understand and represent cell functionality, and they This, in turn, provides a suitable framework for versatile,
have repeatedly yielded meaningful biological insights sophisticated analyses that have already been used suc-
[12]. However, although the standard network paradigm cessfully to reveal multilayer community structure [14]
has been very successful, it has a fundamental flaw: it and to measure important nodes and the correlations be-
forces the aggregation of multilayer information to con- tween them [15–17]. However, to meet the requirements
struct network representations that include only a single of an operational toolbox to be applied to the analysis
type of connection between pairs of entities. This can of biological systems, it is of paramount importance to
lead to lead to misleading results, and it is becoming also develop a system to visualize multilayer networks
increasingly apparent that a more complicated represen- and represent the results of analysis in a meaningful way.
tation is necessary (see the review by Kivela et al [13] The primary contributions of the present work are to
and references therein). address the computational challenge of analysis and visu-
The increasing use of more complicated network rep- alization of multilayer information by providing a prac-
resentations has yielded a new set of challenges: how tical methodology, and accompanying software that we
should one visualize, analyze, and interpret multilayer call muxViz , for the analysis and the visualization of the
biological data. For instance, in a recent study, the ge- salient features of multilayer networks. In Supplemen-
netic and protein-protein interaction networks of Sac- tary Notes 5 and 6, we give technical details about the
charomyces cerevisiae were investigated simultaneously muxViz software and its graphical user interface, together
[11] to uncover connection patterns. In another exam- with a few representative examples of analysis of multi-
ple, Costanzo et al [11] reported that genetic interac- layer biological networks (see Supplementary Note 4). In
tions have an overlap of 10–20% with protein-protein multilayer networks, nodes can exist on several layers si-
interaction pairs, which is significantly higher than the multaneously and counterpart nodes from different layers
3% overlap that they expected based on a random null are connected to each other via inter-layer edges. One can
model. This suggests that many positive and negative visualize a multilayer network in muxViz either using ex-
interactions occur between — rather than within — com- plicit layers or as an edge-colored multigraph [13, 18], in
plexes and pathways [11] and thereby gives an important which edges are “colored” according to the different type
example of how exploiting multilayer information might of relationships between them (see Supplementary Fig-
improve understanding of biological structure and func- ure 1 for applications to genetic and neuronal multilayer
tionality. networks).
 2

To demonstrate the ability of muxViz to analyze and represents the merging process as a “reducibility dendro-
visualize multilayer networks, let’s consider different gram” like the one in Figure 1C. The muxViz software
types of genetic interactions for organisms in the Bio- controls this procedure via a quality function that guar-
logical General Repository for Interaction Datasets [19] antees the merging of redundant layers with minimum
(BioGRID, thebiogrid.org), a public database that loss of information with respect to the full multilayer rep-
archives and disseminates genetic and protein interac- resentation (see Supplementary Note 2). Naturally, one
tion data from humans and model organisms. BioGRID can also use other ways of preserving information in such
currently includes more than 720,000 interactions that a reduction process.
have been curated from both high-throughput data sets In Figure 1D, we show degree-degree Spearman corre-
and individual focused studies using over 41,000 publica- lation coefficients between layers to quantify the tendency
tions in the primary literature. We use BioGRID 3.2.108 of nodes to be hubs in different layers simultaneously.
(updated 1 Jan 2014). In the present paper, we focus on The muxViz software also includes additional correlation
Xenopus laevis and show a network visualization in Fig- measures (see Supplementary Note 4), and it is easy for
ure 1A. We give results of computations using muxViz in users to implement other indicators [17].
the other panels of Figure 1. See Supplementary Note 4 To summarize all of the information that one obtains
for the full analysis of this and other organisms. from calculations like the ones above in a compact figure,
One can examine the global organization of nodes into we developed an annular visualization that facilitates the
modules (i.e., “communities”) through an algorithmic ability to capture patterns to deduce qualitative infor-
calculation of community structure [14]. For example, mation about multilayer data. In Figure 1E, we show
one can obtain dense communities in multilayer networks an example for centrality diagnostics, which measure the
by optimizing a multilayer generalization of the modu- importance of nodes in various ways. Each ring indicates
larity quality function [14]. To do this, one takes into a centrality measure, and the angle determines the iden-
account both intra-layer and inter-layer edges, and one tity of a node in a network, regardless of the layer(s) in
seeks densely connected sets of nodes (i.e., communities) which it exists. In this visualization, we have binned the
that are sparsely connected to each other as compared centrality values, and the color indicates the value. To
to some multilayer random-graph model [14]. See Sup- maximize the readability of the annular plots, we adopted
plementary Note 4 for a visualization of communities in several criteria are adopted (see Supplementary Note 3),
Xenopus laevis and other organisms. although users are free to choose custom options. One
One can quantify the importance of a node by using can use the same principles when fixing some centrality
various diagnostics to measure “centrality”. One calcu- descriptor and letting the rings correspond to the layers
lates such a centrality (and a corresponding rank order) in a network, the multilayer network, and an aggregated
for each node by using multilayer generalizations of cen- network (see Figure 1F, where we consider strength cen-
trality measures [13, 15, 16]. The software muxViz has trality). For the case of layers, one calculates centrality
tools for calculating several different types of centrality for each layer separately without accounting for multi-
(e.g., degree, eigenvector, hub and authority, PageRank, layer structure. For instance, it is evident that rings
and Katz) either for an entire multilayer network or for 3 (“DirInt” layer) and 5 (“PhAssoc” layer) are nega-
each layer separately. As we illustrate in Figure 1B, cen- tively correlated because nodes tend to have opposite
trality values (as well as other network measures) can be colors, whereas rings 6 (aggregated network) and 7 (mul-
very different in multilayer networks than in their corre- tiplex network) are positively correlated, as expected for
sponding aggregations. Such results influence how one strength centrality. Our annular representation makes
should interpret calculations of network measures for, it easy to see similarity (or dissimilarity) in rank or-
e.g., which genes or proteins are most important for ac- derings according to different diagnostics. For example,
tivating or suppressing a given biological processes. The their patterns reveal that physical association and direct
data in question is multilayer, so the analysis of such data interaction are dominant and determine the multilayer
must take multilayer features into account. strength. In other cases (see Supplementary Note 4), the
Researchers are often also interested in considering a ranking by some centrality measure in the multilayer net-
“reduced version” of multilayer data sets that preserve work is poorly correlated to the ranking in either an ag-
as much information as possible without altering the pri- gregated network or in individual layers separately. This
mary descriptors. For such scenarios, it is possible to underscores the importance of using a multilayer frame-
use dimensionality reduction to identify the layers of a work for the calculation of the most central proteins.
multilayer network that are providing redundant infor- In the current era of “big data”, there is now an in-
mation [20]. For example, one can calculate a pairwise tense deluge of multilayer data. To avoid throwing away
distance between layers and can in turn hierarchically important information or obtaining misleading results, it
cluster the layers using this distance. (As explained in is increasingly crucial to use methods that exploit mul-
Supplementary Note 2, users can choose among several tilayer structure. In this paper, we present new soft-
clustering methods.) One then merges the layers and ware and associated methodology that exploits the new
 3

Figure 1: Multilayer analysis of genetic interaction data. (A) A three-dimensional edge-colored multigraph represen-
tation of empirical relationships in Xenopus laevis. (B) Multilayer representation and the corresponding aggregated network,
where we label the five nodes with the highest values of multilayer PageRank centrality [16]. (C) Distance matrix, based
on quantum Jensen-Shannon divergence between each pair of layers and the corresponding reducibility dendrogram (see Sup-
plementary Note 2), which indicates the order in which pairs of layers are combined in hierarchical clustering [20]. (D)
Degree-degree correlations quantified by pairwise Spearman coefficients between layers. (E) Annular visualization (see Sup-
plementary Note 3) of multilayer diagnostics; each ring represents a type of centrality. (F) Annular visualization of strength
centrality: rings represent the layers, the multilayer network, or the corresponding aggregation. (In panels (E,F), we give the
labels on the upper right from the inner ring (top) to the outer ring (bottom).
 4

paradigm of multilayer networks, and we illustrate how multiscale, and multiplex networks. Science 328, 876–
it can be used to analyze and visualize multi-relational 878 (2010).
genetic networks. Our software, muxViz , provides an [15] De Domenico, M. et al. Mathematical formulation of
open-source framework for multilayer analysis. Addition- multilayer networks. Phys. Rev. X 3, 041022 (2013).
[16] De Domenico, M., Solé-Ribalta, A., Omodei, E., Gómez,
ally, the modular structure of muxViz — along with its S. & Arenas, A. Centrality in interconnected multilayer
open-source license — makes it easy to add new methods networks. arXiv:1311.2906 (2013).
for analysis. Moreover, although we have illustrated the [17] Nicosia, V. & Latora, V. Measuring and modelling cor-
power of muxViz for the analysis of biological networks, relations in multiplex networks. arXiv:1403.1546 (2014).
it clearly is also useful for multilayer networks from any [18] Nicosia, V., Bianconi, G., Latora, V. & Barthelemy,
other setting. As we illustrate in Supplementary Fig- M. Growing multiplex networks. Phys. Rev. Lett. 111,
ure 10, it can even be overlaid over spatial information. 058701 (2013).
[19] Stark, C. et al. Biogrid: a general repository for inter-
All authors were supported by the European Com- action datasets. Nucleic Acids Research 34, D535–D539
mission FET-Proactive project PLEXMATH (Grant No. (2006).
317614). AA also acknowledges financial support from [20] De Domenico, M., Nicosia, V., Arenas, A. & Latora, V.
the Generalitat de Catalunya 2009-SGR-838, the ICREA Layer aggregation and reducibility of multilayer intercon-
Academia, and the James S. McDonnell Foundation. nected networks. arXiv:1405.0425 (2014).
MAP acknowledges a grant (EP/J001759/1) from the [21] Newman, M. E. J. Networks: An Introduction (Oxford
University Press, 2010).
EPSRC. We thank Serafina Agnello for her support with
[22] Jeong, H., Tombor, B., Albert, R., Oltvai, Z. N. &
graphics. Barabási, A.-L. The large-scale organization of metabolic
networks. Nature 407, 651–654 (2000).
[23] Jeong, H., Mason, S. P., Barabási, A.-L. & Oltvai, Z. N.
References Lethality and centrality in protein networks. Nature 411,
41–42 (2001).
[24] Maslov, S. & Sneppen, K. Specificity and stability in
topology of protein networks. Science 296, 910–913
[1] Kitano, H. Computational systems biology. Nature 420, (2002).
206–210 (2002). [25] Guimerà, R. & Amaral, L. A. N. Functional cartography
[2] De Jong, H. Modeling and simulation of genetic regula- of complex metabolic networks. Nature 433, 895–900
tory systems: a literature review. Journal of computa- (2005).
tional biology 9, 67–103 (2002). [26] Goh, K.-I. et al. The human disease network. Proc. Natl.
[3] Barabási, A.-L. & Oltvai, Z. N. Network biology: under- Acad. Sci. USA 104, 8685–8690 (2007).
standing the cell’s functional organization. Nature Re- [27] van Vreeswijk, C. & Sompolinsky, H. Chaos in neuronal
views Genetics 5, 101–113 (2004). networks with balanced excitatory and inhibitory activ-
[4] Sharan, R., Ulitsky, I. & Shamir, R. Network-based pre- ity. Science 274, 1724–1726 (1996).
diction of protein function. Molecular Systems Biology 3 [28] Sporns, O. & Kötter, R. Motifs in brain networks. PLoS
(2007). biology 2, e369 (2004).
[5] Beyer, A., Bandyopadhyay, S. & Ideker, T. Integrating [29] Buzsáki, G. & Draguhn, A. Neuronal oscillations in cor-
physical and genetic maps: from genomes to interaction tical networks. Science 304, 1926–1929 (2004).
networks. Nature Reviews Genetics 8, 699–710 (2007). [30] Sporns, O., Chialvo, D. R., Kaiser, M. & Hilgetag,
[6] Sporns, O. Contributions and challenges for network C. C. Organization, development and function of com-
models in cognitive neuroscience. Nature Neuroscience plex brain networks. Trends in cognitive sciences 8, 418–
(2014). 425 (2004).
[7] Shen-Orr, S. S., Milo, R., Mangan, S. & Alon, U. Net- [31] Mantini, D., Perrucci, M. G., Del Gratta, C., Romani,
work motifs in the transcriptional regulation network of G. L. & Corbetta, M. Electrophysiological signatures of
Escherichia coli. Nature genetics 31, 64–68 (2002). resting state networks in the human brain. PNAS 104,
[8] Tong, A. H. Y. et al. Global mapping of the yeast genetic 13170–13175 (2007).
interaction network. Science 303, 808–813 (2004). [32] Bullmore, E. & Sporns, O. Complex brain networks:
[9] Rosenfeld, N., Young, J. W., Alon, U., Swain, P. S. & graph theoretical analysis of structural and functional
Elowitz, M. B. Gene regulation at the single-cell level. systems. Nature Reviews Neuroscience 10, 186–198
Science 307, 1962–1965 (2005). (2009).
[10] Chen, B. L., Hall, D. H. & Chklovskii, D. B. Wiring [33] Seeley, W. W., Crawford, R. K., Zhou, J., Miller, B. L. &
optimization can relate neuronal structure and function. Greicius, M. D. Neurodegenerative diseases target large-
PNAS 103, 4723–4728 (2006). scale human brain networks. Neuron 62, 42–52 (2009).
[11] Costanzo, M. et al. The genetic landscape of a cell. Sci- [34] Bassett, D. S. et al. Dynamic reconfiguration of human
ence 327, 425–431 (2010). brain networks during learning. PNAS 108, 7641–7646
[12] Nature methods. In Supplement on visualizing biological (2011).
data, vol. 7 (Nature Publishing Group, 2010). [35] Nicosia, V., Valencia, M., Chavez, M., Dı́az-Guilera, A. &
[13] Kivelä, M. et al. Multilayer networks. Journal of Complex Latora, V. Remote synchronization reveals network sym-
Networks, In press. arXiv:1309.7233 (2014). metries and functional modules. Phys. Rev. Lett. 110,
[14] Mucha, P. J., Richardson, T., Macon, K., Porter, M. A. 174102 (2013).
& Onnela, J.-P. Community structure in time-dependent, [36] Nicosia, V., Vértes, P. E., Schafer, W. R., Latora, V.
 5

& Bullmore, E. T. Phase transition in the economically SUPPLEMENTARY MATERIAL

modeled growth of a cellular nervous system. PNAS 110,
7880–7885 (2013). Supplementary Note 1: Multilayer Networks in
[37] Buldyrev, S. V., Parshani, R., Paul, G., Stanley, H. E. muxViz
& Havlin, S. Catastrophic cascade of failures in interde-
pendent networks. Nature 464, 1025–1028 (2010).
[38] Brummitt, C. D., D’Souza, R. M. & Leicht, E. Suppress- In the last two decades, the analysis of complex sys-
ing cascades of load in interdependent networks. PNAS tems has benefited significantly from the modeling in-
109, E680–E689 (2012). teractions and relationships using networks [21]. This
[39] Radicchi, F. & Arenas, A. Abrupt transition in the has led to significant improvements in scientific under-
structural formation of interconnected networks. Nature
standing of numerous phenomena — including genetic
Physics 9, 717–720 (2013).
[40] Battiston, F., Nicosia, V. & Latora, V. Structural mea- and metabolic processes [7–9, 11, 22–26], neuronal con-
sures for multiplex networks. Phys. Rev. E 89, 032804 nectivity of living beings [27–36], and more.
(2014). However, despite these many successes, the framework
[41] Yağan, O. & Gligor, V. Analysis of complex contagions of ordinary networks—which encompass only one type
in random multiplex networks. Phys. Rev. E 86, 036103 of connection between each pair of entities and which
(2012). are concerned with static snapshots or aggregations of
[42] De Domenico, M., Solé-Ribalta, A., Gómez, S. &
networks—leaves out crucial information that can lead
Arenas, A. Random walks on multiplex networks.
arXiv:1306.0519 (2013). to misleading or even incorrect results. As a result, the
[43] DeLellis, P., di Bernardo, M., Gorochowski, T. E. & much richer framework of multilayer networks [13] has
Russo, G. Synchronization and control of complex net- attracted an ever-increasing amount of interest among
works via contraction, adaptation and evolution. Circuits scientists. Multilayer networks make it possible to ana-
and Systems Magazine, IEEE 10, 64–82 (2010). lyze much more complicated systems in a realistic way, as
[44] Bassett, D. S. et al. Robust detection of dynamic com- now it is possible to encompass both temporal dynam-
munity structure in networks. Chaos 23, 013142 (2013).
ics of relationships and multiple types of relationships
[45] Fruchterman, T. M. & Reingold, E. M. Graph draw-
ing by force-directed placement. Software: Practice and in a natural way so that one can throw away much less
experience 21, 1129–1164 (1991). information than is required in the usual network frame-
[46] Kamada, T. & Kawai, S. An algorithm for drawing gen- work. Multilayer networks have already yielded fascinat-
eral undirected graphs. Information processing letters 31, ing insights and are experiencing burgeoning popularity
7–15 (1989). [13]. For example, there have been numerous studies
[47] Majtey, A., Lamberti, P. & Prato, D. Jensen-shannon to attempt to understand how interdependencies (e.g.,
divergence as a measure of distinguishability between
[37, 38]), multilayer structures (e.g., [39, 40]), dynam-
mixed quantum states. Phys. Rev. A 72, 052310 (2005).
[48] Braunstein, S. L., Ghosh, S. & Severini, S. The lapla- ics (e.g., [41, 42]), and control (e.g., [43]) can improve
cian of a graph as a density matrix: a basic combinato- understanding of complex interacting systems. See the
rial approach to separability of mixed states. Annals of recent review article [13] for extensive discussions and a
Combinatorics 10, 291–317 (2006). thorough review of results.
[49] Briët, J. & Harremoës, P. Properties of classical and The muxViz software focuses predominantly on mul-
quantum jensen-shannon divergence. Phys. Rev. A 79, tiplex networks, which refer to networks with multiple
052311 (2009).
types of relational types and are arguably the most im-
[50] Gan, G., Ma, C. & Wu, J. Data Clustering: Theory,
Algorithms, and Applications (SIAM, 2007). portant (and prevalent) type of multilayer network. A
[51] Cardillo, A. et al. Emergence of network features from large variety of real-world systems in the biological, so-
multiplexity. Scientific Reports 3, 1344 (2013). cial, information, physical, and engineering sciences can
[52] Lima, A., De Domenico, M., Pejovic, V. & Musolesi, be described as multiplex networks. In such a network, a
M. Exploiting cellular data for disease containment and node can exist on different layers simultaneously, and a
information campaigns strategies in country-wide epi- node in a given layer is connected to its counterparts
demics. arXiv:1306.4534 (2013).
on other layers via inter-layer edges. In muxViz , we
consider two different types of inter-layer connectivity:
ordinal and categorical. In ordinal multilayer networks,
inter-layer edges exist only between layers that are ad-
jacent to each other with respect to some criterion (e.g.,
temporal ordering). By contrast, categorical multilayer
networks include inter-layer edges between counterpart
nodes from every pair of layers. For the sake of simplic-
ity, we illustrate muxViz using inter-layer edges of weight
1 in this paper. In general, how to choose such weights is
an open research question. See the discussion in Ref. [44].
For instance, let’s examine the genetic-interaction and
 6

Figure 1: Supplementary Figure 1: Multilayer representations of genetic and neuronal networks. (A) Multilayer
representation, in which the layers correspond to interaction network of genes in Saccharomyces cerevisiae (which was obtained
via a synthetic genetic-array methodology) and a correlation-based network in which genes with similar interaction profiles are
connected to each other. [The data comes from Ref. [11].] In the third later, we show the corresponding aggregated network.
In this visualization, the color of the nodes is their module assignment from multilayer community detection (see the text for
further details). (B). Representation of the same network as an edge-colored multigraph. (C) Multilayer and (D) edge-colored-
multigraph representations of the Caenorhabditis elegans connectome, where layers correspond to different synaptic junctions:
electric (“ElectrJ”), chemical monadic (“MonoSyn”), and polyadic (“PolySyn”). [The data comes from Ref. [10].] In panels B
and D, we color the nodes according to the layer to which they belong. If a node is part of multiple layers simultaneously, then
we use an equal distribution of the corresponding colors for the node.

profile-correlation networks of a cell, which were aggre- nodes by combining two of the standard force-directed
gated into a single network in Ref [11], as different layers algorithms available in muxViz and applying them to
of a multilayer network. In Supplementary Figure 1A, an aggregated network that we obtained by summing
we show multilayer visualizations that we created using the corresponding entries of the adjacency matrices of
muxViz . Other representations are also possible [13]. the individual layers. Specifically, we first apply the
For example, when representing this data as an edge- Fruchterman-Reingold algorithm [45] to the aggregated
colored multigraph, we “color” edges according to the network and then use the output of this algorithm as
type of relationship that they represent (see Supplemen- a seed layout for the Kamada-Kawai algorithm [46] to
tary Figure 1B). In Supplementary Figure 1C and Sup- achieve a better spatial separation of nodes in the fi-
plementary Figure 1D, we show the two visualizations for nal layout. The muxViz software also allows other layout
the connectome of Caenorhabditis elegans. In this exam- choices: for example, the layout of each layer can be in-
ple, each layer corresponds to a different type of synaptic dependent, or one can use any individual or aggregation
connection [10]. of any subset of layers to determine node locations.

In panels (A) and (C) of Supplementary Figure 1, we

use a layout in which the positions of the nodes are
the same in each layer. We determine the positions of
 7

Supplementary Note 2: Dimensionality reduction

and reducibility dendrogram

An important open question is the determination of

how much information is necessary to accurately rep-
resent the structure of multilayer systems and whether
some layers can be aggregated without incurring loss
of information. It was shown recently that it is pos-
sible to reduce the number of layers in multilayer net-
works in a way that minimizes information loss by using
an information-theoretical approach [20]. The method-
ology of Ref. [20], which we implemented in muxViz ,
amounts to a tradeoff (which is “optimal” in some sense)
between accuracy and complexity. Alternatively, users
of muxViz can implement alternative methods based on
different interpretations of “minimal loss information”.
The reduction procedure from Ref. [20] proceeds as fol-
lows. For each pair of layers in the original multilayer net-
work, muxViz calculates the quantum Jensen–Shannon
divergence [47]. This estimates the similarity between
two networks based on their Von Neumann entropy [48].
By definition, the quantum Jensen–Shannon divergence
is symmetric and its square root, which is usually called
the Jensen–Shannon distance, satisfies the properties of a
metric [49]. The JS distance can be used to quantify the
distance in terms of information gain/loss between the
normalized Laplacian matrices associated to two distinct
networks [20].
One places the distances between every pair of layers
as the components of a matrix, so one can then perform
hierarchical clustering [50] using any desired method to
produce a dendrogram that indicates the relatedness of
the information in the different layers. In muxViz , we
have included several methods for hierarchical cluster-
ing (e.g., Ward, McQuitty, single, complete, average, me-
dian, and centroid linkage). We show an example of such
a “reducibility dendrogram” in panel (D) of Supplemen-
tary Figure 3. A reducibility dendrogram merges a set
of layers in a multilayer network step by step, and we
calculate a quality function based on the relative Von
Neumann entropy to estimate information gain (or loss)
at each step [20]. To obtain a reduced version of the orig-
inal multilayer network, we stop the merging procedure
at the level of the hierarchy that maximizes the relative
entropy.
 8

Supplementary Note 3: Annular visualization of distances, which we then hierarchically cluster to group
multivariate information the rings. This clustering procedure determines the or-
der of the rings to try to maximize the readability of the
It is a challenging problem to represent, visualize, and annular plot.
analyze the wealth of information encoded in the mul- One can also use the same principles when fixing some
tilayer structure of networks in a compact way. Pre- centrality descriptor and letting the rings correspond to
serving more information by using multilayer networks the layers in a network, the multilayer network, and an
rather than ordinary networks then complicates the vi- aggregated network. Such a plot might help to reveal, for
sualization and analysis even further. However, this com- instance, if the centrality of nodes in a multilayer network
plication is necessary, because otherwise one might end is primarily due to their centrality in a specific layer or if
up with misleading or even incorrect results [13]. We de- the aggregated network is a good proxy for the multilayer
veloped the muxViz software to help address these chal- structure.
lenges. To summarize all of the information obtained
from multilayer-network calculations in a compact figure,
muxViz includes an annular visualization that facilitates
the ability to capture patterns and deduce qualitative
information about multilayer data.
To give a concrete example, many researchers are in-
terested in ranking the relative importance of nodes (and
other network structures), which traditionally is accom-
plished using various “centrality” measures. Centralities
have been calculated for single-layer networks for several
decades, and numerous notions of centrality are now also
available for multilayer networks [13, 16]. It is there-
fore necessary to develop visualization tools that make it
possible to compare such a wealth of diagnostics to each
other in a compact, meaningful way. For example, it is
often of interest to focus attention on one descriptor and
compare the values obtained in each layer separately to
the values obtained from the multilayer network and its
aggregations. The muxViz software makes it easy to do
this.
We will now illustrate the annular visualization (e.g.,
see Supplementary Figure 5) using the example of multi-
layer centrality measures. Suppose that we have different
arrays of information, where you should think of each
array as having resulted from the calculation of some
centrality diagnostic on a multilayer network. We vi-
sualize each array in a ring. The angle indicates node
identity (regardless of the layer or layers in which it oc-
curs). We bin the centrality values—e.g., either linearly
or logarithmically—and we assign a color to each bin
to encode its value. Both the type of binning and the
color scheme are customizable in muxViz . We place the
rings concentrically, and one can determine both their
order and their thicknesses according to any desired cri-
teria. For example, in the visualizations in the present
paper, we determine the thickness of each ring according
to its information content (which we quantify using the
Shannon information entropy of the distribution of the
values): thinner rings have less information. Users can
customize the order of the rings, but as a default it is de-
termined automatically via hierarchically clustering. The
muxViz software calculates a measure of correlation (e.g.,
Pearson, Spearman, or Jensen–Shannon divergence) be-
tween each pairs of descriptors to obtain a set of pairwise
 9

Supplementary Note 4: Example analyses of In the second set of figures (see Supplementary Fig-
empirical multilayer networks ures 3, 5, 7 and 9), we show the annular visualization for
the centrality descriptors:
In this note, we present multilayer analyses of four bi-
ological systems to illustrate the power of muxViz . We • In panels titled “Multiplex”, we consider the multi-
examine the following examples: layer network. Each ring corresponds to a different
centrality descriptor.
• Xenopus laevis genetic-interaction network (see
Supplementary Figures 2 and 3); • In the other panels, we consider a specific central-
ity descriptor (which we specify in the title of the
• Caenorhabditis elegans connectome (see Supple- panel). Each ring encodes the values of that de-
mentary Figures 4 and 5); scriptor, which we calculate separately in each layer
separately. We also include rings for the calcula-
• Herpes simplex genetic-interaction network (see
tion of the corresponding centrality diagnostic in
Supplementary Figures 6 and 7);
the multilayer network and in its aggregation to a
• HIV-1 genetic-interaction network (see Supple- single-layer weighted network.
mentary Figures 8 and 9).
We specify the order of the rings in the list of labels on
We include two figures for each example. In the first the right of each plot. In each case (and as in Figure 1
set of figures (see Supplementary Figures 2, 4, 6, and 8), in the main text), the top label refers to the innermost
we show the following information: ring and the bottom label refers to the outermost ring.

• Panel A: Multilayer community structure from

modularity maximization [14]. The color of each
node encodes its community assignment in a
multilayer-network visualization. For comparison,
we also show the results (and corresponding visual-
ization) of community detection on an aggregated
network, which we obtain by summing the corre-
sponding intra-layer edge weights of all layers. (In
other words, if Aijs gives the edge weight between
nodes i and j on layer s, then we obtain an aggre-
gated edge weight Wij between nodes i and j by
summing over s.)
• Panel B: Multilayer centrality [16]. We again use a
multilayer-network visualization. We label the top
five nodes from a ranking according to multilayer
PageRank centrality. For comparison, we also show
the results of centrality calculations on the afore-
mentioned aggregated network.
• Panel C: Edge-colored multigraph visualization of
the network. We color edges according to the layer
to which they belong. We color the nodes according
to their layer (or layers); if a node exists on multiple
layers, then we distribute its corresponding colors
evenly.
• Panel D: Dimensionality-reduction analysis and
corresponding reducibility dendrogram [20]. We
show the distance matrix and the corresponding
dendrogram, which we obtain using Ward hierar-
chical clustering.
• Panel E: Measures of correlation between layers:
(left) mean edge overlap, (center) degree-degree
Pearson correlation coefficient, and (right) degree-
degree Spearman correlation coefficient.
 10

Figure 2: Supplementary Figure 2: Multilayer analysis of a Xenopus laevis genetic-interaction network. See the
text of Supplementary Note 4 for details about each panel. [In this figure and all subsequent figures, we have purposely kept
font sizes at muxViz ’s default level rather than increasing them.]
 11

Figure 3: Supplementary Figure 3: Multilayer analysis of a Xenopus laevis genetic-interaction network. See the
text of Supplementary Note 4 for details about each panel.
 12

Figure 4: Supplementary Figure 4: Multilayer analysis of a Caenorhabditis elegans connectome. See the text of
Supplementary Note 4 for details about each panel.
 13

Figure 5: Supplementary Figure 5: Multilayer analysis of a Caenorhabditis elegans connectome. See the text of
Supplementary Note 4 for details about each panel.
 14

Figure 6: Supplementary Figure 6: Multilayer analysis of a Herpes simplex genetic-interaction network. See the
text of Supplementary Note 4 for details about each panel.
 15

Figure 7: Supplementary Figure 7: Multilayer analysis of a Herpes simplex genetic-interaction network. See the
text of Supplementary Note 4 for details about each panel. Note that we do not show eigenvector centrality because one layer
consists of a directed acyclic graph (for which eigenvector centrality is unilluminating [21]).
 16

Figure 8: Supplementary Figure 8: Multilayer analysis of HIV-1 genetic-interaction network. See the text of
Supplementary Note 4 for details about each panel.
 17

Figure 9: Supplementary Figure 9: Multilayer analysis of HIV-1 genetic-interaction network. See the text of
Supplementary Note 4 for details about each panel.
 18

Supplementary Note 5: Technical Details About

muxViz

We developed muxViz using R (http://www.

r-project.org/), a free and widely adopted
framework for statistical computing, and GNU Oc-
tave (https://www.gnu.org/software/octave/),
an open-source high-level interpreted language
that is intended primarily for numerical com-
putations. The Octave language is very sim-
ilar to the proprietary environment Matlab
(http://www.mathworks.es/products/matlab/),
and one can import the code to Matlab in a straight-
forward manner. The muxViz software requires R 3.0.2
(or above) and Octave 3.4.0 (or above).
The muxViz framework is a free and open-source pack-
age for the analysis and the visualization of multilayer
networks. It is released under GNU General Public Li-
cense v3 (https://www.gnu.org/copyleft/gpl.html)
and exploits R to provide an easy and accessible user in-
terface for the visualization of networks, the calculation
of network diagnostics, and the visual representation of
the results of calculations. Specifically, R allows the con-
struction of a graphical user interface (GUI), which can
be used either locally (client-side software) or via the in-
ternet (remote Web server), and an Octave library that
we developed performs calculations of matrices and ten-
sors.
Using muxViz is simple and does not require any pro-
gramming skill; one can do all computations and visu-
alization via the user interface. Additionally, because of
muxViz ’s modular structure, users can also create their
own modules for calculating new diagnostics and for cus-
tomizing visual representations.
The muxViz framework allows both two-dimensional
and three-dimensional visualization of networks. The
latter exploits OpenGL technology, so users can interac-
tively change the perspective and navigate the network.
We show representative static snapshots of such interac-
tive visualizations in Supplementary Figures 1B and D
and in panel C of Supplementary Figures 2, 4, 6, and 8.
Obviously, muxViz is not limited to multilayer biologi-
cal data, as it can be used on multilayer networks in gen-
eral. One can even incorporate spatial information into
the visualization. In Supplementary Figure 10, for exam-
ple, we include geographical information to help visualize
networks of (A) European airports and (B) districts in
the Ivory Coast.
 19

Figure 10: Supplementary Figure 10: Multilayer networks embedded in geographical regions. (A) Network of
European airports, where each layer represents a different airline [51]. (B) Network of mobility and communication in Ivory
Coast, where nodes are geographical districts [52]. We used muxViz to visualize these data sets.
 20

Supplementary Note 6: Graphical User Interface tions for network visualization and for graphical represen-
tation of the results. We show screenshots of muxViz in
The graphical user interface (GUI) of muxViz allows Supplementary Figures 11, 12, and 13. In Supplemen-
one to use many methods and diagnostics for the investi- tary Figure 14, we include a table that summarizes some
gation of multilayer networks. It also includes many op- centrality diagnostics.
 21

Figure 11: Supplementary Figure 11: muxViz Graphical User Interface. Primary panel of muxViz graphical user
interface to import networks, calculate diagnostics (e.g., centrality and correlation measures), visualize networks, and perform
dimensionality reduction on the data.

Figure 12: Supplementary Figure 12: muxViz GUI. The diagnostics panel of muxViz allows users to set up the diagnostics
(and their corresponding parameters) to be calculated. Users also access the annular visualization from this panel.
 22

Figure 13: Supplementary Figure 13: muxViz GUI. Some of the options that users can customize to visualize multilayer
networks. Several different layout algorithms are available, and there are also several graphical options to manipulate the
appearance of layers, nodes, and edges.
 23

Figure 14: Supplementary Figure 14: muxViz GUI. Example of diagnostics summarized in a table by muxViz .

View publication stats