Scalable Computing of Betweenness Centrality based on Graph Reduction with a Case Study on Breast Cancer Analytics

Abstract

BackgroundGraph theory has been widely applied to the studies in biomedicine such as structural measures including betweenness centrality. However, if the network size is too large, the result of betweenness centrality would be difficult to obtain in a reasonable amount of time.ResultIn this paper, we describe an approach, 1+ɛ lossy graph reduction algorithm, to computing betweenness centrality on large graphs. The approach is able to guarantee a bounded approximation result. We use GSE48216, a breast cancer cell line co-expression network, to show that our algorithms can achieve a higher reduction rate with a trade-off of some bounded errors in query results. Furthermore, by comparing the betweenness centrality of the original graph and the reduced graph, it can be shown that a higher reduction rate does not sacrifice the accuracy of betweenness centrality when providing faster execution time.ConclusionsOur proposed 1+ɛ lossy graph reduction algorithm is validated by the experiment results which show that the approach achieves a faster execution within a bounded error rate.

An Experimental Study on Centrality Measures Using Clustering

“…Their algorithm takes a set of features for each node and the adjacency matrix as its input and uses them to estimate the centrality rank of each node. In [19], they propose a lossy graph reduction approach that reduces the execution time of the centrality algorithms. After our investigations in this field, we decided to extend our previous research [20] and investigate the ranking efficiency and execution time of more centrality measures using clustering methods.…”