Community detection in complex networks using extremal optimization

Enhanced modularity-based community detection by random walk network preprocessing

The representation of real systems with network models is becoming increasingly common and critical to both capture and simplify systems' complexity, notably, via the partitioning of networks into communities. In this respect, the definition of modularity, a common and broadly used quality measure for networks partitioning, has induced a surge of efficient modularity-based community detection algorithms. However, recently, the optimization of modularity has been found to show a resolution limit, which reduces its effectiveness and range of applications. Therefore, one recent trend in this area of research has been related to the definition of novel quality functions, alternative to modularity. In this paper, however, instead of laying aside the important body of knowledge developed so far for modularity-based algorithms, we propose to use a strategy to preprocess networks before feeding them into modularity-based algorithms. This approach is based on the observation that dynamic processes triggered on vertices in the same community possess similar behavior patterns but dissimilar on vertices in different communities. Validations on real-world and synthetic networks demonstrate that network preprocessing can enhance the modularity-based community detection algorithms to find more natural clusters and effectively alleviates the problem of resolution limit.

Loops and multiple edges in modularity maximization of networks

The modularity maximization model proposed by Newman and Girvan for the identification of communities in networks works for general graphs possibly with loops and multiple edges. However, the applications usually correspond to simple graphs. These graphs are compared to a null model where the degree distribution is maintained but edges are placed at random. Therefore, in this null model there will be loops and possibly multiple edges. Sharp bounds on the expected number of loops, and their impact on the modularity, are derived. Then, building upon the work of Massen and Doye, but using algebra rather than simulation, we propose modified null models associated with graphs without loops but with multiple edges, graphs with loops but without multiple edges and graphs without loops nor multiple edges. We validate our models by using the exact algorithm for clique partitioning of Grötschel and Wakabayashi.

Statistical significance of communities in networks

Nodes in real-world networks are usually organized in local modules. These groups, called communities, are intuitively defined as subgraphs with a larger density of internal connections than of external links. In this work, we define a measure aimed at quantifying the statistical significance of single communities. Extreme and order statistics are used to predict the statistics associated with individual clusters in random graphs. These distributions allows us to define one community significance as the probability that a generic clustering algorithm finds such a group in a random graph. The method is successfully applied in the case of real-world networks for the evaluation of the significance of their communities.

Performance of modularity maximization in practical contexts

Although widely used in practice, the behavior and accuracy of the popular module identification technique called modularity maximization is not well understood in practical contexts. Here, we present a broad characterization of its performance in such situations. First, we revisit and clarify the resolution limit phenomenon for modularity maximization. Second, we show that the modularity function Q exhibits extreme degeneracies: it typically admits an exponential number of distinct high-scoring solutions and typically lacks a clear global maximum. Third, we derive the limiting behavior of the maximum modularity Qmax for one model of infinitely modular networks, showing that it depends strongly both on the size of the network and on the number of modules it contains. Finally, using three real-world metabolic networks as examples, we show that the degenerate solutions can fundamentally disagree on many, but not all, partition properties such as the composition of the largest modules and the distribution of module sizes. These results imply that the output of any modularity maximization procedure should be interpreted cautiously in scientific contexts. They also explain why many heuristics are often successful at finding high-scoring partitions in practice and why different heuristics can disagree on the modular structure of the same network. We conclude by discussing avenues for mitigating some of these behaviors, such as combining information from many degenerate solutions or using generative models.

Edge ratio and community structure in networks

A hierarchical divisive algorithm is proposed for identifying communities in complex networks. To that effect, the definition of community in the weak sense of Radicchi [Proc. Natl. Acad. Sci. U.S.A. 101, 2658 (2004)] is extended into a criterion for a bipartition to be optimal: one seeks to maximize the minimum for both classes of the bipartition of the ratio of inner edges to cut edges. A mathematical program is used within a dichotomous search to do this in an optimal way for each bipartition. This includes an exact solution of the problem of detecting indivisible communities. The resulting hierarchical divisive algorithm is compared with exact modularity maximization on both artificial and real world data sets. For two problems of the former kind optimal solutions are found; for five problems of the latter kind the edge ratio algorithm always appears to be competitive. Moreover, it provides additional information in several cases, notably through the use of the dendrogram summarizing the resolution. Finally, both algorithms are compared on reduced versions of the data sets of Girvan and Newman [Proc. Natl. Acad. Sci. U.S.A. 99, 7821 (2002)] and of Lancichinetti [Phys. Rev. E 78, 046110 (2008)]. Results for these instances appear to be comparable.

Spectral properties of networks with community structure

In this paper, we discuss the eigenspectra of networks with community structure. It is shown that in many cases, the spectrum of eigenvalues of the adjacency matrix of a network with community structure gives a clear indication of the number of communities in the network. In particular, for a network with N nodes and N_(c) communities, there will typically be N_(c) eigenvalues that are significantly larger than the magnitudes of all the other (N-N_(c)) eigenvalues. We discuss this property as well as its use and limitations for determining N_(c) .

Weak signal transmission in complex networks and its application in detecting connectivity

We present a network model of coupled oscillators to study how a weak signal is transmitted in complex networks. Through both theoretical analysis and numerical simulations, we find that the response of other nodes to the weak signal decays exponentially with their topological distance to the signal source and the coupling strength between two neighboring nodes can be figured out by the responses. This finding can be conveniently used to detect the topology of unknown network, such as the degree distribution, clustering coefficient and community structure, etc., by repeatedly choosing different nodes as the signal source. Through four typical networks, i.e., the regular one dimensional, small world, random, and scale-free networks, we show that the features of network can be approximately given by investigating many fewer nodes than the network size, thus our approach to detect the topology of unknown network may be efficient in practical situations with large network size.

Community detection in networks with positive and negative links

Detecting communities in complex networks accurately is a prime challenge, preceding further analyses of network characteristics and dynamics. Until now, community detection took into account only positively valued links, while many actual networks also feature negative links. We extend an existing Potts model to incorporate negative links as well, resulting in a method similar to the clustering of signed graphs, as dealt with in social balance theory, but more general. To illustrate our method, we applied it to a network of international alliances and disputes. Using data from 1993-2001, it turns out that the world can be divided into six power blocs similar to Huntington's civilizations, with some notable exceptions.

Iterated tabu search for identifying community structure in complex networks

This paper presents an iterated tabu search (denoted by ITS) algorithm for optimizing the modularity of community structure in complex networks. The proposed algorithm follows a general framework composed of two phases: basic optimization and postrefinement. When the basic optimization cannot improve the modularity any more, a postrefinement procedure is employed to further optimize the objective function with a global view. For both these two phases, iterated tabu search algorithm is employed to optimize the objective function. Computational results show the high effectiveness of the proposed ITS algorithm compared with six state-of-the-art algorithms in the literature. In particular, our ITS algorithm improves the previous best known modularity for several small and medium size networks.

Analysis of community structure in networks of correlated data

We present a reformulation of modularity that allows the analysis of the community structure in networks of correlated data. The modularity preserves the probabilistic semantics of the original definition even when the network is directed, weighted, signed, and has self-loops. This is the most general condition one can find in the study of any network, in particular those defined from correlated data. We apply our results to a real network of correlated data between stores in the city of Lyon (France).

Alternative approach to community detection in networks

The problem of community detection is relevant in many disciplines of science and modularity optimization is the widely accepted method for this purpose. It has recently been shown that this approach presents a resolution limit by which it is not possible to detect communities with sizes smaller than a threshold, which depends on the network size. Moreover, it might happen that the communities resulting from such an approach do not satisfy the usual qualitative definition of commune; i.e., nodes in a commune are more connected among themselves than to nodes outside the commune. In this paper we present a different method for community detection in complex networks. We define merit factors based on the weak and strong community definitions formulated by Radicchi [Proc. Natl. Acad. Sci. U.S.A. 101, 2658 (2004)] and we show that these local definitions avoid the resolution limit problem found in the modularity optimization approach.

Modulated modularity clustering as an exploratory tool for functional genomic inference

In recent years, the advent of high-throughput assays, coupled with their diminishing cost, has facilitated a systems approach to biology. As a consequence, massive amounts of data are currently being generated, requiring efficient methodology aimed at the reduction of scale. Whole-genome transcriptional profiling is a standard component of systems-level analyses, and to reduce scale and improve inference clustering genes is common. Since clustering is often the first step toward generating hypotheses, cluster quality is critical. Conversely, because the validation of cluster-driven hypotheses is indirect, it is critical that quality clusters not be obtained by subjective means. In this paper, we present a new objective-based clustering method and demonstrate that it yields high-quality results. Our method, modulated modularity clustering (MMC), seeks community structure in graphical data. MMC modulates the connection strengths of edges in a weighted graph to maximize an objective function (called modularity) that quantifies community structure. The result of this maximization is a clustering through which tightly-connected groups of vertices emerge. Our application is to systems genetics, and we quantitatively compare MMC both to the hierarchical clustering method most commonly employed and to three popular spectral clustering approaches. We further validate MMC through analyses of human and Drosophila melanogaster expression data, demonstrating that the clusters we obtain are biologically meaningful. We show MMC to be effective and suitable to applications of large scale. In light of these features, we advocate MMC as a standard tool for exploration and hypothesis generation.

Simple probabilistic algorithm for detecting community structure

With the growing number of available social and biological networks, the problem of detecting the network community structure is becoming more and more important which acts as the first step to analyze these data. The community structure is generally regarded as that nodes in the same community tend to have more edges and less if they are in different communities. We propose a simple probabilistic algorithm for detecting community structure which employs expectation-maximization (SPAEM). We also give a criterion based on the minimum description length to identify the optimal number of communities. SPAEM can detect overlapping nodes and handle weighted networks. It turns out to be powerful and effective by testing simulation data and some widely known data sets.

Community structures and role detection in music networks

We analyze the existence of community structures in two different social networks using data obtained from similarity and collaborative features between musical artists. Our analysis reveals some characteristic organizational patterns and provides information about the driving forces behind the growth of the networks. In the similarity network, we find a strong correlation between clusters of artists and musical genres. On the other hand, the collaboration network shows two different kinds of communities: rather small structures related to music bands and geographic zones, and much bigger communities built upon collaborative clusters with a high number of participants related through the period the artists were active. Finally, we detect the leading artists inside their corresponding communities and analyze their roles in the network by looking at a few topological properties of the nodes.

Partition of networks into basins of attraction

We study partition of networks into basins of attraction based on a steepest ascent search for the node of highest degree. Each node is associated with, or "attracted" to its neighbor of maximal degree, as long as the degree is increasing. A node that has no neighbors of higher degree is a peak, attracting all the nodes in its basin. Maximally random scale-free networks exhibit different behavior based on their degree distribution exponent gamma : For small gamma (broad distribution) networks are dominated by a giant basin, whereas for large gamma (narrow distribution) there are numerous basins, with peaks attracting mainly their nearest neighbors. We derive expressions for the first two moments of the number of basins. We also obtain the complete distribution of basin sizes for a class of hierarchical deterministic scale-free networks that resemble random nets. Finally, we generalize the problem to regular networks and lattices where all degrees are equal, and thus the attractiveness of a node must be determined by an assigned weight, rather than the degree. We derive the complete distribution of basins of attraction resulting from randomly assigned weights in one-dimensional chains.

Automatic decomposition of kinetic models of signaling networks minimizing the retroactivity among modules

MOTIVATION

The modularity of biochemical networks in general, and signaling networks in particular, has been extensively studied over the past few years. It has been proposed to be a useful property to analyze signaling networks: by decomposing the network into subsystems, more manageable units are obtained that are easier to analyze. While many powerful algorithms are available to identify modules in protein interaction networks, less attention has been paid to signaling networks de.ned as chemical systems. Such a decomposition would be very useful as most quantitative models are de.ned using the latter, more detailed formalism.

RESULTS

Here, we introduce a novel method to decompose biochemical networks into modules so that the bidirectional (retroactive) couplings among the modules are minimized. Our approach adapts a method to detect community structures, and applies it to the so-called retroactivity matrix that characterizes the couplings of the network. Only the structure of the network, e.g. in SBML format, is required. Furthermore, the modularized models can be loaded into ProMoT, a modeling tool which supports modular modeling. This allows visualization of the models, exploiting their modularity and easy generation of models of one or several modules for further analysis. The method is applied to several relevant cases, including an entangled model of the EGF-induced MAPK cascade and a comprehensive model of EGF signaling, demonstrating its ability to uncover meaningful modules. Our approach can thus help to analyze large networks, especially when little a priori knowledge on the structure of the network is available.

AVAILABILITY

The decomposition algorithms implemented in MATLAB (Mathworks, Inc.) are freely available upon request. ProMoT is freely available at http://www.mpi-magdeburg.mpg.de/projects/promot.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Benchmark graphs for testing community detection algorithms

Community structure is one of the most important features of real networks and reveals the internal organization of the nodes. Many algorithms have been proposed but the crucial issue of testing, i.e., the question of how good an algorithm is, with respect to others, is still open. Standard tests include the analysis of simple artificial graphs with a built-in community structure, that the algorithm has to recover. However, the special graphs adopted in actual tests have a structure that does not reflect the real properties of nodes and communities found in real networks. Here we introduce a class of benchmark graphs, that account for the heterogeneity in the distributions of node degrees and of community sizes. We use this benchmark to test two popular methods of community detection, modularity optimization, and Potts model clustering. The results show that the benchmark poses a much more severe test to algorithms than standard benchmarks, revealing limits that may not be apparent at a first analysis.

Adaptive clustering algorithm for community detection in complex networks

Community structure is common in various real-world networks; methods or algorithms for detecting such communities in complex networks have attracted great attention in recent years. We introduced a different adaptive clustering algorithm capable of extracting modules from complex networks with considerable accuracy and robustness. In this approach, each node in a network acts as an autonomous agent demonstrating flocking behavior where vertices always travel toward their preferable neighboring groups. An optimal modular structure can emerge from a collection of these active nodes during a self-organization process where vertices constantly regroup. In addition, we show that our algorithm appears advantageous over other competing methods (e.g., the Newman-fast algorithm) through intensive evaluation. The applications in three real-world networks demonstrate the superiority of our algorithm to find communities that are parallel with the appropriate organization in reality.

Multistep greedy algorithm identifies community structure in real-world and computer-generated networks

We have recently introduced a multistep extension of the greedy algorithm for modularity optimization. The extension is based on the idea that merging l pairs of communities (l>1) at each iteration prevents premature condensation into few large communities. Here, an empirical formula is presented for the choice of the step width l that generates partitions with (close to) optimal modularity for 17 real-world and 1100 computer-generated networks. Furthermore, an in-depth analysis of the communities of two real-world networks (the metabolic network of the bacterium E. coli and the graph of coappearing words in the titles of papers coauthored by Martin Karplus) provides evidence that the partition obtained by the multistep greedy algorithm is superior to the one generated by the original greedy algorithm not only with respect to modularity, but also according to objective criteria. In other words, the multistep extension of the greedy algorithm reduces the danger of getting trapped in local optima of modularity and generates more reasonable partitions.

Communicability and multipartite structures in complex networks at negative absolute temperatures

We here present a method of clearly identifying multipartite subgraphs in a network. The method is based on a recently introduced concept of the communicability, which very clearly identifies communities in a complex network. We here show that, while the communicability at a positive temperature is useful in identifying communities, the communicability at a negative temperature is useful in identifying multipartite subgraphs; the latter quantity between two nodes is positive when the two nodes belong to the same subgraph and is negative when they do not. The method is able to discover "almost" multipartite structures, where intercommunity connections vastly outweigh intracommunity connections. We illustrate the relevance of this work to real-life food web and protein-protein interaction networks.

Comparative definition of community and corresponding identifying algorithm

A comparative definition for community in networks is proposed, and the corresponding detecting algorithm is given. A community is defined as a set of nodes, which satisfies the requirement that each node's degree inside the community should not be smaller than the node's degree toward any other community. In the algorithm, the attractive force of a community to a node is defined as the connections between them. Then employing an attractive-force-based self-organizing process, without any extra parameter, the best communities can be detected. Several artificial and real-world networks, including the Zachary karate club, college football, and large scientific collaboration networks, are analyzed. The algorithm works well in detecting communities, and it also gives a nice description of network division and group formation.

Community detection by signaling on complex networks

Based on a signaling process of complex networks, a method for identification of community structure is proposed. For a network with n nodes, every node is assumed to be a system which can send, receive, and record signals. Each node is taken as the initial signal source to excite the whole network one time. Then the source node is associated with an n -dimensional vector which records the effects of the signaling process. By this process, the topological relationship of nodes on the network could be transferred into a geometrical structure of vectors in n -dimensional Euclidean space. Then the best partition of groups is determined by F statistics and the final community structure is given by the K -means clustering method. This method can detect community structure both in unweighted and weighted networks. It has been applied to ad hoc networks and some real networks such as the Zachary karate club network and football team network. The results indicate that the algorithm based on the signaling process works well.

Optimal partition and effective dynamics of complex networks

Given a large and complex network, we would like to find the best partition of this network into a small number of clusters. This question has been addressed in many different ways. Here we propose a strategy along the lines of optimal prediction for the Markov chains associated with the dynamics on these networks. We develop the necessary ingredients for such an optimal partition strategy, and we compare our strategy with the previous ones. We show that when the Markov chain is lumpable, we recover the partition with respect to which the chain is lumpable. We also discuss the case of well-clustered networks. Finally, we illustrate our strategy on several examples.

Differential equations as a tool for community identification

We consider the task of identification of a cluster structure in random networks. The results of two methods are presented: (i) the Newman algorithm [M. E. J. Newman and M. Girvan, Phys. Rev. E 69, 026113 (2004)]; and (ii) our method based on differential equations. A series of computer experiments is performed to check if in applying these methods we are able to determine the structure of the network. The trial networks consist initially of well-defined clusters and are disturbed by introducing noise into their connectivity matrices. Further, we show that an improvement of the previous version of our method is possible by an appropriate choice of the threshold parameter beta . With this change, the results obtained by the two methods above are similar, and our method works better, for all the computer experiments we have done.

Bicomponents and the robustness of networks to failure

We study bicomponents in networks, sets of nodes such that each pair in the set is connected by at least two independent paths, so that the failure of no single node in the network can cause them to become disconnected. We show that standard network models predict there to be essentially no small bicomponents in most networks, but there may be a giant bicomponent, whose presence coincides with the presence of the ordinary giant component, and we find that real networks seem by and large to follow this pattern, although there are some interesting exceptions. We also study the size of the giant bicomponent as nodes in the network fail and find in some cases that our networks are quite robust to failure, with large bicomponents persisting until almost all vertices have been removed.

Mesoscopic structure conditions the emergence of cooperation on social networks

Background

We study the evolutionary Prisoner's Dilemma on two social networks substrates obtained from actual relational data.

Methodology/Principal Findings

We find very different cooperation levels on each of them that cannot be easily understood in terms of global statistical properties of both networks. We claim that the result can be understood at the mesoscopic scale, by studying the community structure of the networks. We explain the dependence of the cooperation level on the temptation parameter in terms of the internal structure of the communities and their interconnections. We then test our results on community-structured, specifically designed artificial networks, finding a good agreement with the observations in both real substrates.

Conclusion

Our results support the conclusion that studies of evolutionary games on model networks and their interpretation in terms of global properties may not be sufficient to study specific, real social systems. Further, the study allows us to define new quantitative parameters that summarize the mesoscopic structure of any network. In addition, the community perspective may be helpful to interpret the origin and behavior of existing networks as well as to design structures that show resilient cooperative behavior.

Efficient modularity optimization by multistep greedy algorithm and vertex mover refinement

Identifying strongly connected substructures in large networks provides insight into their coarse-grained organization. Several approaches based on the optimization of a quality function, e.g., the modularity, have been proposed. We present here a multistep extension of the greedy algorithm (MSG) that allows the merging of more than one pair of communities at each iteration step. The essential idea is to prevent the premature condensation into few large communities. Upon convergence of the MSG a simple refinement procedure called "vertex mover" (VM) is used for reassigning vertices to neighboring communities to improve the final modularity value. With an appropriate choice of the step width, the combined MSG-VM algorithm is able to find solutions of higher modularity than those reported previously. The multistep extension does not alter the scaling of computational cost of the greedy algorithm.

Robustness of community structure in networks

The discovery of community structure is a common challenge in the analysis of network data. Many methods have been proposed for finding community structure, but few have been proposed for determining whether the structure found is statistically significant or whether, conversely, it could have arisen purely as a result of chance. In this paper we show that the significance of community structure can be effectively quantified by measuring its robustness to small perturbations in network structure. We propose a suitable method for perturbing networks and a measure of the resulting change in community structure and use them to assess the significance of community structure in a variety of networks, both real and computer generated.

Quantitative function for community detection

We propose a quantitative function for community partition -- i.e., modularity density or D value. We demonstrate that this quantitative function is superior to the widely used modularity Q and also prove its equivalence with the objective function of the kernel k means. Both theoretical and numerical results show that optimizing the new criterion not only can resolve detailed modules that existing approaches cannot achieve, but also can correctly identify the number of communities.

Identifying network communities with a high resolution

Community structure is an important property of complex networks. The automatic discovery of such structure is a fundamental task in many disciplines, including sociology, biology, engineering, and computer science. Recently, several community discovery algorithms have been proposed based on the optimization of a modularity function (Q) . However, the problem of modularity optimization is NP-hard and the existing approaches often suffer from a prohibitively long running time or poor quality. Furthermore, it has been recently pointed out that algorithms based on optimizing Q will have a resolution limit; i.e., communities below a certain scale may not be detected. In this research, we first propose an efficient heuristic algorithm QCUT, which combines spectral graph partitioning and local search to optimize Q . Using both synthetic and real networks, we show that QCUT can find higher modularities and is more scalable than the existing algorithms. Furthermore, using QCUT as an essential component, we propose a recursive algorithm HQCUT to solve the resolution limit problem. We show that HQCUT can successfully detect communities at a much finer scale or with a higher accuracy than the existing algorithms. We also discuss two possible reasons that can cause the resolution limit problem and provide a method to distinguish them. Finally, we apply QCUT and HQCUT to study a protein-protein interaction network and show that the combination of the two algorithms can reveal interesting biological results that may be otherwise undetected.

Modularity and community detection in bipartite networks

The modularity of a network quantifies the extent, relative to a null model network, to which vertices cluster into community groups. We define a null model appropriate for bipartite networks, and use it to define a bipartite modularity. The bipartite modularity is presented in terms of a modularity matrix B; some key properties of the eigenspectrum of B are identified and used to describe an algorithm for identifying modules in bipartite networks. The algorithm is based on the idea that the modules in the two parts of the network are dependent, with each part mutually being used to induce the vertices for the other part into the modules. We apply the algorithm to real-world network data, showing that the algorithm successfully identifies the modular structure of bipartite networks.

Modular networks emerge from multiconstraint optimization

Modular structure is ubiquitous among complex networks. We note that most such systems are subject to multiple structural and functional constraints, e.g., minimizing the average path length and the total number of links, while maximizing robustness against perturbations in node activity. We show that the optimal networks satisfying these three constraints are characterized by the existence of multiple subnetworks (modules) sparsely connected to each other. In addition, these modules have distinct hubs, resulting in an overall heterogeneous degree distribution.

Near linear time algorithm to detect community structures in large-scale networks

Community detection and analysis is an important methodology for understanding the organization of various real-world networks and has applications in problems as diverse as consensus formation in social communities or the identification of functional modules in biochemical networks. Currently used algorithms that identify the community structures in large-scale real-world networks require a priori information such as the number and sizes of communities or are computationally expensive. In this paper we investigate a simple label propagation algorithm that uses the network structure alone as its guide and requires neither optimization of a predefined objective function nor prior information about the communities. In our algorithm every node is initialized with a unique label and at every step each node adopts the label that most of its neighbors currently have. In this iterative process densely connected groups of nodes form a consensus on a unique label to form communities. We validate the algorithm by applying it to networks whose community structures are known. We also demonstrate that the algorithm takes an almost linear time and hence it is computationally less expensive than what was possible so far.

Module identification in bipartite and directed networks

Modularity is one of the most prominent properties of real-world complex networks. Here, we address the issue of module identification in two important classes of networks: bipartite networks and directed unipartite networks. Nodes in bipartite networks are divided into two nonoverlapping sets, and the links must have one end node from each set. Directed unipartite networks only have one type of node, but links have an origin and an end. We show that directed unipartite networks can be conveniently represented as bipartite networks for module identification purposes. We report on an approach especially suited for module detection in bipartite networks, and we define a set of random networks that enable us to validate the approach.