A Robust Density-Based Hierarchical Clustering Algorithm

Abstract

Clustering the genes based on their expression patterns is one of the important subjects in analyzing microarray data. Discovering the genes co-expressed in particular conditions has been done by different clustering algorithms. In these methods, the similar genes are located in the same cluster. Thus, the closer the similar genes, the further the dissimilar ones will be. Each of the applied methods to discover gene clusters has had advantages and drawbacks. The proposed method, which is density-based hierarchical, is robust enough due to discovering clusters with different shapes and detecting noise. Moreover, its hierarchical characteristic illustrates a proper image of data distribution and their relationships. In this paper, the results obtained from executing the algorithm for 30 times show it has notable accuracy to capture clusters, in a way that it is 98% for extracting three-cluster gene networks and 70% for four-cluster ones.