Abstract
Many dynamic systems display complex emergent phenomena. By directly controlling a subset of system components (nodes) via external intervention it is possible to indirectly control every other component in the system. When the system is linear or can be approximated sufficiently well by a linear model, methods exist to identify the number and connectivity of a minimum set of external inputs (constituting a socalled minimal control topology, or MCT). In general, many MCTs exist for a given network; here we characterize a broad ensemble of empirical networks in terms of the fraction of nodes and edges that are always, sometimes, or never a part of an MCT. We study the relationships between the measures, and apply the methodology to the TLGL leukemia signaling network as a case study. We show that the properties introduced in this report can be used to predict key components of biological networks, with potentially broad applications to network medicine.
Introduction
A system of interacting components can be represented by a network, where system components are represented by nodes (or vertices) and their interactions are represented by directed edges (or links) between nodes. The dynamical behavior of the system is constrained by the topology of the network, but, because an edge generally indicates the existence of a regulatory relationship without specifying its nature, network topology is necessarily an incomplete representation of the system’s dynamical behavior. For instance, the dynamic relationship represented by a node A being regulated by two nodes B and C may be wellapproximated by a truth table (e.g., the future state of A is given by B AND C, where the node state variables are binary, 0 or 1) or by an arbitrarily complex function of B and C (e.g., dA/dt = B^{C}, where the node state variables are real numbers).
Whatever the dynamical relationships, a common goal is to efficiently influence the behavior of the constituent complex system. For example, manipulating biological systems can potentially lead to the development of effective therapies for disease^{1,2,3,4,5}, ecological management may preserve vital ecosystem services^{6,7,8,9,10}, and technological systems such as the power grid must be carefully designed and managed to provide functional and efficient services to consumers^{11,12,13,14,15}. The field of network science aims to identify the extent to which the network structure alone can be used to study the dynamics, and more recently the control, of systems. In part because such analysis aims to identify modelagnostic properties and in part because it can provide informative firstorder effects, it is common to model the evolution of systems by linear dynamics of the form
where x is a vector of the nodes in the network, A is the adjacency matrix that encodes the interactions in the network topology (including, for instance, diagonal terms corresponding to the free decay of a node state variable), and the term B u encodes external, usercontrolled signals fed into a subset of socalled directlycontrolled nodes, i.e., the dynamics of these specific nodes are directly affected by these external signals. Many systems, of course, are inherently nonlinear and are at best approximately linear within a local region; assuming linear dynamics of the form of equation (1) in these cases is an extremely coarse approximation^{16,17,18}, thought note that linearizing a nonlinear system is in some cases a sufficient condition for local controllability^{19}. Nonetheless, the study of linear dynamics is important not only for systems that are linear, but also as a comparative framework for nonlinear dynamics: a system is nonlinear to the extent that it departs from linear expectations. The extent to which our results apply to nonlinear systems will depend on the form of the nonlinearity and the region of interest. While the networks we study behave according to a variety of models, we use their interaction networks to provide realistic structures for us to study the potential degeneracies of control properties.
In this case it is possible, with a sufficient number and placement of directlycontrolled nodes and appropriate choices of timevarying control signals, to drive the system to any desired state in finite time^{20,21,22,23}. In principle one wishes to control the dynamics of a network with as few interventions as possible; minimizing the number of controls amounts to finding the maximum matching on the network (see Methods). Recent work has addressed the relationship between network topology and the properties of the directlycontrolled nodes. For instance, the directlycontrolled nodes are either source nodes or arise due to a dilation, where a node has more than one outgoing edge^{23}. By classifying dilations as external (if the outgoing edges point to sink nodes) or internal (otherwise), one can determine the fraction of controls in each of these categories (respectively denoted η_{s}, η_{e}, and η_{i}). These parameters constitute the control profile of a network. Empirical networks tend to be dominated by one of the three control profile parameters; diverse mechanisms exist by which synthetic networks may be generated with the same properties^{23,24,25,26}.
While the control profile offers insight into the types of directlycontrolled nodes in terms of their topological location, it offers no direct insight into the degeneracy of control: there are generally many solutions to the maximum matching of a network. In other words, many combinations of nodes may be chosen for direct control. Furthermore, determining the maximum matching involves not only a set of directlycontrolled nodes, but also a set of matched edges, which constitute control signal paths (see Fig. 1, Methods). It is therefore of interest to assess the importance of both nodes and edges in the set of all of the socalled minimal control topologies (MCTs). Specifically, here we wish to determine if a node is always, sometimes, or never directly controlled among all maximum matchings (i.e., across all MCTs). Similarly, we wish to determine if an edge is always, sometimes, or never on a control signal path.
Our contributions in this report are threefold. First, we leverage the methodology of Jia et al.^{21} to characterize a new networklevel statistic capturing the fraction of nodes that are always, sometimes, or never directly controlled (υ_{a}, υ_{s}, and υ_{n}, respectively). We then develop a new method using a similar approach to assign a networklevel statistic capturing the fraction of edges that always, sometimes, or never belong to the set of matched edges (fractions ε_{a}, ε_{s}, and ε_{n}, respectively). These new quantities echo the spirit of the control profile, which is a unique networklevel statistic that captures the fraction of controls due to different functional structures. In contrast, these new quantities capture the fraction of nodes and edges that participate in the process of control signal dissemination through the network in varying ways. Thus the third contribution of this work is to provide insight into the connections between the functional classification of the control configuration (given by the control profile) and the degeneracy of the control configuration allowed by the network structure (captured by the statistics proposed in this report). Notably, we show that the existence of many internal dilations correlates with many edges never existing on a control path, suggesting that internal dilations tend to restrict the flexibility with which control signals propagate through a network. Finally, we apply the methodology of this report to the TLGL leukemia signaling network and show broad agreement between the metrics introduced herein and existing experimental, computational, and analytical work that has identified nodes whose control play a pivotal role in the behavior of the network. While (as noted above) it is known that the study of structural controllability has some applications to systems characterized by nonlinear dynamics, the abovesummarized results provide additional evidence suggesting that the methodology described in this report may be used in other biological networks to predict network components essential for control, with potentially broad applications to network medicine.
Results
We analyze 58 empirical networks (see Table S1) and determine their distribution in parameter space for the control profile, nodebased degeneracy, and edgebased degeneracy measures (Fig. 2a–c). As previously reported, the control profile of any one empirical network tends to be dominated by one of the control profile parameters^{23}. In contrast, the nodebased degeneracy measures indicate that ν_{a} ≪ 1 for most networks, that is, few networks have a significant fraction of nodes that are always directly controlled (in agreement with the observation that most networks have relatively few directlycontrolled nodes). This observation applies also to the edgebased degeneracy measures; ε_{a} ≪ 1. However, while networks are welldispersed between ν_{s} and ν_{n}, most networks are skewed toward ε_{s}, meaning that while some networks have many nodes that are never directly controlled, few networks have many edges that are never on a control signal path.
We are also interested in the relationships between these measures. In Fig. 2(d–f) we show the distributions of degeneracy measures separately for networks where each of the control profile measures is largest (e.g., η_{s} > η_{e}, η_{i}). We find, for instance, that η_{s}dominated networks tend to be ν_{n} and ε_{s}dominated. In other words, networks where many of the directlycontrolled nodes are source nodes tend to have many nodes that are never directly controlled, and many edges that can be (but are not necessarily) on control signal paths. In contrast, η_{e}dominated networks tend to be ν_{s} and ε_{s}dominated, meaning that networks with an abundance of sink nodes tend to have significant flexibility both in choice of directlycontrolled nodes and control signal paths. Finally, η_{i}dominated networks tend to be ε_{s}dominated (though less so than in the case of either η_{s}dominance or η_{e}dominance) and distributed between ν_{s} and ν_{n}. Thus, networks with many internal dilations may or may not have flexibility in terms of choice of directlycontrolled nodes, and tend to have at least a moderate degree of flexibility in control signal paths.
We perform a pairwise Spearman correlation analysis between each of the 9 controlrelated parameters considered here and with basic network properties (Table 1). Unsurprisingly, there is a strong negative correlation between ε_{a} and each of the average node degree, average clustering coefficient, and network transitivity (−0.79, −0.52, and −0.58, respectively): more connections per node and/or an increased frequency of closed triads afford greater flexibility in assigning control signal paths. Interestingly, these same properties are negatively correlated with ν_{s} and positively correlated with ν_{n}: a richer local structure constrains nodal participation (as directly controlled nodes) in MCTs. Furthermore, the number of nodes and edges exhibit weak negative correlation with ν_{n} and ε_{n}, indicating that larger networks are more likely to access more nodes and edges in at least some MCTs. Measures within a set tend to be negatively correlated with one another, with the notable exception of ε_{a}and ε_{n} (0.29). Some powerful trends exist between sets, as well, as suggested by Fig. 2: η_{s} is correlated with ν_{a} (0.8), η_{i} is correlated with ε_{n} (0.74), and η_{e} is correlated with ν_{s} (0.68).
To validate the utility of these measures, we consider the dynamic model of survival signaling network relevant to TLGL leukemia^{2,27}. In this disease a fraction of white blood cells activated in response to a stimulus escape the process of activation induced cell death, survive, and after a while start attacking healthy cells. The dynamics of this network are defined by Boolean functions, from which a topological network can be extracted such that A>B if node A exists in the update function for node B. In the TLGL network the node representing apoptosis (i.e., programmed cell death) is of particular interest. Its OFF state, together with the deregulation (abnormally high or low activity) of a subset of nodes, indicates the abnormal, leukemic state. Conversely, if in a leukemic cell the state of apoptosis cell changes from OFF to ON, the cell is committed to the process of cell death. Existing work has identified the minimal set of nodes whose sustained expression can lead to the leukemic state. This set consists of three source nodes: the initial stimulus, as well as the external molecules plateletderived growth factor (PDGF) and interleukin (IL) 15, both of which were experimentally observed to be overabundant in the blood of TLGL leukemia patients^{27}. Prior work has also identified nodes whose direct control can lead to apoptosis of leukemic cells, despite the continued presence of these source nodes. Control of any one of 18 nodes (of 57 total) leads to apoptosis according to at least two of the following three types of evidence: experimental verification^{27}, simulation of Boolean dynamics^{2}, and analysis of the topology of the network once it has been expanded to topologically encode the Boolean rules^{28}.
The extent to which we expect the present metrics to agree with prior work is mitigated to some extent by the scope of the methodologies: in most prior work the quantity of interest is the state of a single node (apoptosis), whereas structural controllability seeks to achieve a desired state for every node in the network. Furthermore, the methodology used here assumes dynamics that obey equation (1), which is quite different from the Boolean framework used in the prior work being considered here. Therefore, a conservative expectation is that the two methodologies do not contradict one another. Specifically, assuming that the target state reflects induced apoptosis of a leukemic cell, we expect that the three source nodes necessary for the leukemic state are always directly controlled and the 18 apoptosisinducing nodes should, at minimum, sometimes be directly controlled and/or be connected to an edge that is always on a control path.
We verify that this is the case: all three source nodes are always directly controlled, and 15 of the 18 key nodes have at least one incoming or outgoing “always” edge (indicating that they take part in a critical signaling pathway in terms of control) and/or are sometimes directly controlled (indicating that in some cases system control may require direct control of these nodes). Furthermore, the remaining three key nodes are connected to at least 4 “sometimes” edges (indicating flexibility in the manner in which control signals are routed through these nodes). Indeed, despite the different methodological frameworks, the agreement is rather strong: all 18 nodes are connected to more “always” and/or more “sometimes” edges than expected by random chance, and only 2 are connected to more “never” edges than expected by random chance (for more details, see the Supplementary Information).
Discussion
Effectively influencing the behavior of complex interacting systems is a broad, multidisciplinary goal. Accordingly, there is significant interest in discovering general techniques by which the dynamics of systems from different domains (e.g., technological and biological) may be guided by external intervention. We here consider systems that obey the linear dynamics of equation (1), where it has been shown that complete control is possible by feeding external control signals into a subset of the system components^{20,21,22,23}. These directlycontrolled nodes are chosen such that every other node in the network is reached via nonoverlapping paths originating at the directlycontrolled nodes (see Fig. 1). The directlycontrolled nodes and these control paths together constitute a control topology; a minimal control topology (MCT) is one in which the number of inputs is minimized.
In this report we consider the degeneracy of minimal control (i.e. the extent to which different minimal control topologies exist for a given network) in linear systems. Specifically, we characterize every system component (node) and interaction (edge) as being always, sometimes, or never on a MCT (the fraction of all nodes in these categories are respectively represented by the parameters ν_{a}, ν_{s}, and ν_{n} for nodes and ε_{a}, ε_{s}, and ε_{n} for edges). We study a broad selection of empirical networks and find that they are generally distributed between ν_{s}and ν_{n} while ν_{a}tends to be small. While we can unambiguously state that nodes are always directly controlled only if they are source nodes^{23}, in all but the simplest networks more precise statements require analysis of the maximummatching problem and/or perturbing a MCT via a breadthfirst search (see Methods). However, the flexibility of control in this framework is reflected by the typically high values of ε_{s}, suggesting that there are generally many ways for control signals to propagate through a network, even if there is relatively little flexibility in the choice of nodes to be directly controlled.
We consider these measures against the fraction of controls that are source nodes, sink nodes, and internal nodes (η_{s,}, η_{e} and η_{i}, respectively), quantities which are fixed for a given network^{23}. The fact that η_{s} is positively correlated with ν_{a} follows from their definitions: η_{s} is the fraction of directlycontrolled nodes that are source nodes, and ν_{a} is the fraction of all nodes that are source nodes (see Methods). The fact that η_{i} is positively correlated with ε_{n}indicates the existence of some rigidity in control signal paths cases where most of the directlycontrolled nodes are neither sources nor sinks. In contrast, the correlation between η_{e} and ν_{s} suggests flexibility when most of the control nodes are sink nodes. In other words, there is flexibility in choosing which sink nodes are directly controlled and which are not, likely in part because there are multiple paths from source nodes to different sink nodes. It is also interesting to note that the correlations between nodebased degeneracy measures and edgebased degeneracy measures tends to be weak (with the exception of ν_{a}and ε_{a}, the correlation magnitudes are uniformly below 0.4), indicating a relative disconnect between the nodebased and edgebased degeneracy measures considered in this report.
While an interesting topic from a strictly theoretical standpoint, characterizing control degeneracy also has significant practical implications. In a biological system, for instance, a particular group of signaling molecules may be implicated in many theoretically viable control strategies. This, in turn, could incentivize the development of (e.g., pharmacological) techniques to influence the molecules in question. Indeed, we have shown broad agreement between the techniques developed here and existing work concerning the dynamics of the TLGL leukemia signaling network, and the techniques described herein could be used to identify potential candidates for regulatory control in other biological networks. In an ecological system, the abundance of a particular group (or groups) of species, for example invasive or endangered species, may be controlled to initiate a cascade of changes in the abundances of other species^{6,7,8,29,30,31,32}. Species implicated in many viable control strategies under an appropriate modeling framework may, therefore, be prime candidates for direct manipulation to effectively manage ecological communities. Regardless of context, in cases where the nature of any nonlinearity is unknown, the methods developed here may provide insight into which components are essential for control.
We observe in this study that there exist meaningful correlations between the degeneracy of the control topology (directly controlled nodes and matched edges) and the functional divisions offered by the control profile. While aggregated statistics of network controllability have offered fruitful insights in the past, moving forward – to understand more precisely how network topology is related to network control – will require knowledge about all the possible control paths that can be used to control a network. Ultimately we aim to provide a clear mapping between the structure of the network and the ability we have to control such a system. Here we have provided a new dimension such that we can use the types of degeneracy exhibited by the control topology along with the dominance of certain types of control structures (given by the control profile) to triangulate more informed inferences on the network structures that are most important for network control.
A tempting avenue for future work is the development of procedures that identify the fraction of control paths that contain a given node or edge. This information would allow the categorical analysis considered in this report to be complemented by analysis on a continuum: nodes in V_{a} are in 100% of all control paths, nodes in V_{n} are in 0%, and nodes in V_{s} are somewhere in between (and similarly for edges). Studying the properties that drive nodes and edges to have comparatively high or low participation in control paths promises to enhance our understanding of the relationship between the structure and controllability of complex systems. In addition, we note that the diverse selection of empirical networks evaluated in this study offers insight into network structures that are independent of context. While this follows related work and avoids sample bias that arises when considering traditional generative models^{23,26,33}, taking a similar approach as this study, but focused upon a particular empirical context (e.g., cellular signaling networks, ecological networks) may offer networkspecific insight.
Methods
Control Topology
We define a control topology as a set of directlycontrolled nodes, N_{d}, and the corresponding control signal paths that yield indirect control over every other node in the network^{20,23}. A control topology is minimal if it additionally minimizes N_{d}. Prior work has generally assessed the properties of a single minimal control topology (MCT) for a given network; a MCT is often obtained via the HopcroftKarp algorithm^{20,23,34,35}. In Fig. 1 we show several control topologies for a simple network.
Nodebased assessment of MCT degeneracy
Because many MCTs generally exist for all but the simplest networks, we wish to characterize the nodes in a network according to the frequency with which they are directly controlled in a MCT. Specifically, a node is always, sometimes, or never directly controlled in a MCT; we denote the set of nodes in these categories as V_{a}, V_{s}, and V_{n}, respectively. Similarly, we denote the size of each set, normalized by the total number of nodes in the network, as ν_{a}, ν_{s}, and ν_{n}.
To categorize the nodes in this way, we adopt the method proposed by Jia et al.^{21}. Suppose a single MCT has been determined, and consider first the set of directlycontrolled nodes N_{d}. Clearly every node n ∈ N_{d} is a member of either V_{a} or V_{s}. Making this distinction is trivial in light of the fact that the set of source nodes is identical to V_{a}^{21}. It follows immediately that the directlycontrolled nodes in the MCT that are not source nodes are members of V_{s}.
It remains only to consider the nodes n ∉ N_{d}. Clearly every such node is a member of either V_{s} or V_{n}. To determine the membership of one such node n_{i}, we force it to be directly controlled: if N_{d} increases as a result, then it immediately follows that no MCT directly controls node n_{i} and therefore n_{i} ∈ V_{n}. Otherwise, n_{i} ∈ V_{s}. We repeat this procedure for all nodes n ∉ N_{d}. It is possible to force a node to be directly controlled by perturbing the original MCT with an algorithmic complexity O(EN) (see SI).
Edgebased control classification
Here we are interested in similarly classifying edges as always, sometimes, or never existing on the path of a control signal. We respectively define the sets of nodes in these categories as E_{a}, E_{s}, and E_{n}, and the normalized sizes of these sets as ε_{a}, ε_{s}, and ε_{n}. As in the case of the nodebased analysis, we begin by applying the HopcroftKarp algorithm to the network in question to determine one MCT. From this MCT we obtain a set of edges on control paths, L_{c} (e.g., edge A>B in Fig. 1a).
Clearly edges l ∈ L_{c} are members of E_{a} or E_{s}. Similarly, edges l ∉ L_{c} are members of E_{s} or E_{n}. In the first case, removing one such edge l_{ji} (denoting an edge from node j to node i) and reevaluating the number of directlycontrolled nodes via the HopcroftKarp algorithm serves to categorize the node: if N_{d} increases, l_{ji} ∈ E_{a}; otherwise l_{ji} ∈ E_{s}. In the second case we may wish to force an edge l_{ji} ∉ L_{c} to be on a control path and similarly reapply the HopcroftKarp algorithm; however, no simple modification to the network guarantees that the HopcroftKarp algorithm will force l_{ji} ∈ L_{c}.
We therefore develop alternatives for both of the above cases; the approach has a complexity of O(E^{2}) (see SI).
Additional Information
How to cite this article: Campbell, C. et al. Correlations in the degeneracy of structurally controllable topologies for networks. Sci. Rep. 7, 46251; doi: 10.1038/srep46251 (2017).
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
 1
Campbell, C. & Albert, R. Stabilization of perturbed Boolean network attractors through compensatory interactions. BMC Syst. Biol. 8, 53 (2014).
 2
Saadatpour, A. et al. Dynamical and Structural Analysis of a T Cell Survival Network Identifies Novel Candidate Therapeutic Targets for Large Granular Lymphocyte Leukemia. PLoS Comput Biol. 7, e1002267 (2011).
 3
Kreeger, P. K. & Lauffenburger, D. A. Cancer systems biology: a network modeling perspective. Carcinogenesis 31, 2–8 (2010).
 4
Steinway, S. N. et al. Network Modeling of TGF Signaling in Hepatocellular Carcinoma EpithelialtoMesenchymal Transition Reveals Joint Sonic Hedgehog and Wnt Pathway Activation. Cancer Res. 74, 5963–5977 (2014).
 5
Huang, S., Ernberg, I. & Kauffman, S. Cancer attractors: A systems view of tumors from a gene network dynamics and developmental perspective. Semin. Cell Dev. Biol. 20, 869–876 (2009).
 6
LaBar, T., Campbell, C., Yang, S., Albert, R. & Shea, K. Restoration of plant–pollinator interaction networks via species translocation. Theor. Ecol. 7, 209–220 (2014).
 7
LaBar, T., Campbell, C., Yang, S., Albert, R. & Shea, K. Global versus local extinction in a network model of plant–pollinator communities. Theor. Ecol. 6, 495–503 (2013).
 8
Levin, S. A. Towards a science of ecological management. Conserv. Ecol. 3, 6 (1999).
 9
Attiwill, P. M. The disturbance of forest ecosystems: the ecological basis for conservative management. For. Ecol. Manag. 63, 247–300 (1994).
 10
Liebhold, A. M. & Tobin, P. C. Population Ecology of Insect Invasions and Their Management*. Annu. Rev. Entomol. 53, 387–408 (2008).
 11
Dobson, I. Complex networks: Synchrony and your morning coffee. Nat. Phys. 9, 133–134 (2013).
 12
Hines, P., Blumsack, S., Cotilla Sanchez, E. & Barrows, C. The Topological and Electrical Structure of Power Grids. In 2010 43rd Hawaii International Conference on System Sciences (HICSS), doi: 10.1109/HICSS.2010.398 1–10 (2010).
 13
Motter, A. E., Myers, S. A., Anghel, M. & Nishikawa, T. Spontaneous synchrony in powergrid networks. Nat. Phys. 9, 191–197 (2013).
 14
Pagani, G. A. & Aiello, M. The Power Grid as a complex network: A survey. Physica A 392, 2688–2700 (2013).
 15
Kinney, R., Crucitti, P., Albert, R. & Latora, V. Modeling cascading failures in the North American power grid. Eur. Phys. J. B 46, 101–107 (2005).
 16
Motter, A. E. Networkcontrology. Chaos Interdiscip. J. Nonlinear Sci. 25, 97621 (2015).
 17
Cornelius, S. P., Kath, W. L. & Motter, A. E. Realistic control of network dynamics. Nat. Commun. 4 (2013).
 18
Gates, A. J. & Rocha, L. M. Control of complex networks requires both structure and dynamics. Sci. Rep. 6, 24456 (2016).
 19
Liu, Y.Y. & Barabási, A.L. Control principles of complex systems. Rev. Mod. Phys. 88 (2016).
 20
Liu, Y.Y., Slotine, J.J. & Barabási, A.L. Controllability of complex networks. Nature 473, 167–173 (2011).
 21
Jia, T. et al. Emergence of bimodality in controlling complex networks. Nat. Commun. 4 (2013).
 22
Pósfai, M., Liu, Y.Y., Slotine, J.J. & Barabási, A.L. Effect of correlations on network controllability. Sci. Rep. 3 (2013).
 23
Ruths, J. & Ruths, D. Control Profiles of Complex Networks. Science 343, 1373–1376 (2014).
 24
Ruths, J. & Ruths, D. Response to Comment on ‘Control profiles of complex networks’. Science 346, 561–561 (2014).
 25
Campbell, C., Shea, K. & Albert, R. Comment on ‘Control profiles of complex networks’. Science 346, 561–561 (2014).
 26
Campbell, C., Ruths, J., Ruths, D., Shea, K. & Albert, R. Topological constraints on network control profiles. Sci. Rep. 5, 18693 (2015).
 27
Zhang, R. et al. Network model of survival signaling in large granular lymphocyte leukemia. Proc. Natl. Acad. Sci. 105, 16308–16313 (2008).
 28
Zañudo, J. G. T. & Albert, R. Cell Fate Reprogramming by Control of Intracellular Network Dynamics. PLOS Comput. Biol. 11, e1004193 (2015).
 29
Campbell, C., Yang, S., Albert, R. & Shea, K. Plantpollinator community network response to species invasion depends on both invader and community characteristics. Oikos, doi: 10.1111/oik.02039 406–413 (2014).
 30
Campbell, C., Yang, S., Shea, K. & Albert, R. Topology of plantpollinator networks that are vulnerable to collapse from species extinction. Phys. Rev. E 86, 21924 (2012).
 31
Barrows, C. W. et al. A framework for monitoring multiplespecies conservation plans. J. Wildl. Manag. 69, 1333–1345 (2005).
 32
Knops, J. M. H. et al. Effects of plant species richness on invasion dynamics, disease outbreaks, insect abundances and diversity. Ecol. Lett. 2, 286–293 (1999).
 33
Pequito, S., Preciado, V. M., Barabási, A.L. & Pappas, G. J. Tradeoffs between driving nodes and timetocontrol in complex networks. Sci. Rep. 7, 39978 (2017).
 34
Hopcroft, J. E. & Karp, R. M. An n^{5/2} Algorithm for Maximum Matchings in Bipartite Graphs. SIAM J. Comput. 2, 225–231 (1973).
 35
Commault, C., Dion, J.M. & van der Woude, J. W. Characterization of generic properties of linear structured systems for efficient computations. Kybernetika 38, 503–520 (2002).
Acknowledgements
The authors gratefully acknowledge Shaun Ramsey for useful discussions during the preparation of this report. This work was supported by NSF Grant DMS1313115.
Author information
Affiliations
Contributions
C.C. and S.A. conceived of the study and performed analysis. All authors analyzed results. C.C. wrote the first draft of the manuscript; J.R., D.R., K.S. and R.A. contributed to revisions.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing financial interests.
Supplementary information
Rights and permissions
This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/
About this article
Cite this article
Campbell, C., Aucott, S., Ruths, J. et al. Correlations in the degeneracy of structurally controllable topologies for networks. Sci Rep 7, 46251 (2017). https://doi.org/10.1038/srep46251
Received:
Accepted:
Published:
Further reading

NeuroClinical Signatures of Language Impairments: A Theoretical Framework for Functiontostructure Mapping in Clinics
Current Topics in Medicinal Chemistry (2020)

Strategies to Enhance Logic ModelingBased Cell LineSpecific Drug Synergy Prediction
Frontiers in Physiology (2020)

Irrelevance of linear controllability to nonlinear dynamical networks
Nature Communications (2019)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.