Graph kernels, hierarchical clustering, and network community structure: experiments and comparative analysis

Zhang, Shihua; Ning, Xue-Mei; Zhang, Xiang-Sun

doi:10.1140/epjb/e2007-00146-y

Cited by 20 publications

(14 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Czekanowski -Dice distance has the maximum value for two nodes with no common neighbours and the minimum value for those with exactly the same neighbours. As a similarity index, Zhang et al (2007) use the diffusion kernel of a network, which can capture a longer-range relation between the nodes than the topological overlap or the Czekanowski -Dice distance.…”

Section: Computing the Node Similaritymentioning

confidence: 99%

ModuleSearch: finding functional modules in a protein–protein interaction network

Cui

Shrestha

Han

2012

Computer Methods in Biomechanics and Biomedical Engineering

View full text Add to dashboard Cite

Many biological processes are performed by a group of proteins rather than by individual proteins. Proteins involved in the same biological process often form a densely connected sub-graph in a protein-protein interaction network. Therefore, finding a dense sub-graph provides useful information to predict the function or protein complex of uncharacterised proteins in the sub-graph. We developed a heuristic algorithm that finds functional modules in a protein-protein interaction network and visualises the modules. The algorithm has been implemented in a platform-independent, standalone program called ModuleSearch. In an interaction network of yeast proteins, ModuleSearch found 366 overlapping modules. Of the modules, 71% have a function shared by more than half the proteins in the module and 58% have a function shared by all proteins in the module. Comparison of ModuleSearch with other programs shows that ModuleSearch finds more sub-graphs than most other programs, yet a higher proportion of the sub-graphs correspond to known functional modules. ModuleSearch and sample data are freely available to academics at http://bclab.inha.ac.kr/ModuleSearch.

show abstract

Section: Computing the Node Similaritymentioning

confidence: 99%

ModuleSearch: finding functional modules in a protein–protein interaction network

Cui

Shrestha

Han

2012

Computer Methods in Biomechanics and Biomedical Engineering

View full text Add to dashboard Cite

show abstract

“…This measure has been tested in two collaborative recommendation tasks [46], but did not perform well in this context. Some recent attempts to define similarity measures on nodes of a graph for clustering purposes are [123,140] (performed in parallel with the present work). In [123], Villa et al use a batch version of the kernel self-organizing map, introduced in [83], based on an exponential diffusion kernel, in order to cluster a medieval social network.…”

Section: Similarity-based Approachesmentioning

confidence: 99%

“…In [123], Villa et al use a batch version of the kernel self-organizing map, introduced in [83], based on an exponential diffusion kernel, in order to cluster a medieval social network. On the other hand, Zhang and coworkers perform in [140] a hierarchical clustering directly from the similarities provided by an exponential diffusion kernel. They illustrate the technique on social networks data, such as the karate club and the football team networks.…”

Section: Similarity-based Approachesmentioning

confidence: 99%

Graph nodes clustering with the sigmoid commute-time kernel: A comparative study

Yen

Fouss

Decaestecker

et al. 2009

Data & Knowledge Engineering

View full text Add to dashboard Cite

“…Li et al [29] investigated community detection based on Potts model and the network's spectral characterization and showed that the local, uniform behavior of spins in Potts model can naturally reveal the hierarchical community structures in networks. In recent years, statistical clustering methods in data mining also have been applied to the analysis of multiscale structure in complex networks [30][31][32]. Some multiresolution modularity methods have been studied that will directly or indirectly modify the definition of modularity by tunable parameters [27].…”

Section: Introductionmentioning

confidence: 99%

Analysis of community-detection methods based on Potts spin model in complex networks

et al. 2015

Can. J. Phys.

View full text Add to dashboard Cite

Detection of community structures in complex networks is a common challenge in the study of complex networks. Recently, various methods have been proposed to discover community structures at different scales. Here, the multiscale methods based on Potts spin model for community detection are described and compared in the analysis of community structures of several networks. We give a critical analysis of the multiscale methods, showing a kind of limitation that the methods may suffer from when the community size difference is very broad, the breakup of (large) communities will appear before the merger of (small) communities disappears. In particular, we give the explicit expressions for the critical points of the merger and breakup of communities and derive the sufficient conditions (in the form of upper limits) that indicate when the Potts model methods suffer from the limitation. We apply the theoretical results to model networks and show that the method using the configuration null model (i.e., a random graph model as comparison that has the same degree distribution as the network under study) may not recover the full structure of the model network, whereas the method using the Erdös-Rényi null model will do so. PACS Nos.: 89.75.Hc, 05.10.-a, 89.75.Fb.Résumé : La détection de structures de communautés des réseaux complexes est un défi fréquent dans l'étude de ces réseaux complexes. Récemment, différentes méthodes ont été proposées pour la structure de communautés des réseaux à différentes échelles. Nous décrivons ici les méthodes multi-échelles basées sur le modèle de spin de Potts pour détecter les communautés et les comparons dans l'analyse des structures de communautés de plusieurs réseaux. Nous offrons une analyse critique des méthodes multi-échelles, montrant une sorte de limitation dont peuvent souffrir ces méthodes lorsque la différence en grosseur des communautés devient grande -l'éclatement des grandes communautés apparaitra avant que la réunion des petites communautés disparaisse -. Plus précisément, nous donnons des expressions explicites pour les points de réunion et d'éclatement des communautés et nous dérivons les conditions suffisantes (sous la forme de limites supérieures) qui indiquent quand les méthodes basées sur le modèle de Potts souffrent de cette limitation. Nous appliquons ces résultats théoriques pour modéliser des réseaux et montrons que la méthode utilisant le modèle nul de configuration (i.e. un modèle à graphes aléatoires comme comparaison qui a le même distribution de degrés que le réseau étudié) peut ne pas pouvoir recouvrer la pleine structure du réseau modèle, alors que la méthode utilisant le modèle nul de Erdös-Rényi le peut. [Traduit par le Rédaction]

show abstract

Graph kernels, hierarchical clustering, and network community structure: experiments and comparative analysis

Cited by 20 publications

References 30 publications

ModuleSearch: finding functional modules in a protein–protein interaction network

ModuleSearch: finding functional modules in a protein–protein interaction network

Graph nodes clustering with the sigmoid commute-time kernel: A comparative study

Analysis of community-detection methods based on Potts spin model in complex networks

Contact Info

Product

Resources

About