Efficiently mining rich subgraphs from vertex-attributed graphs

Abstract

With the rapid collection of large network data such as biological networks and social networks, it has become very important to develop efficient techniques for network analysis. In many domains, additional attribute data can be associated with entities and relationships in the network, where the network data represents relationships among entities in the network and the attribute data represents various characteristics of the corresponding entities and relationships in the network. Simultaneous analysis of both network and attribute data results in detection of subnetworks that are contextually meaningful. We propose an efficient algorithm for enumerating all connected vertex sets in an undirected graph. Extending this enumeration approach, an algorithm for enumerating all maximal cohesive connected vertex sets in a vertex-attributed graph is proposed. To prune search branches that will not yield maximal patterns, we also present three pruning techniques for efficient enumeration of the maximal cohesive connected vertex sets. Our comparative runtime analyses show the efficiency and effectiveness of our proposed approaches. Gene set enrichment analysis shows that protein-protein interaction subnetworks with similar cancer dysregulation attributes are biologically significant. Availability: The implementation of the algorithm is available at http://www.cs.ndsu.nodak.edu/~ssalem/richsubgraphs.html