DENCH: A density-based hierarchical clustering algorithm for gene expression data

According to the characteristics of gene expression data, a density-based hierarchical clustering algorithm called DENCH (DENsity-based hierarCHical clustering) is proposed. By defining density for points, DENCH can find arbitrary-shaped clusters and filter noises effectively. Compared with traditional density-based clustering algorithms, DENCH provides both the local view and the global view of the data distribution. Particularly, DENCH arranges the points in a hierarchical framework which overcomes the limit of the global density threshold in traditional density-based clustering algorithms. The basic unit in the hierarchical framework is a set of points rather than one point, which also makes DENCH robust to noises and outliers. DENCH also provides the mechanism to transform the hierarchical structure into directly divided clusters, and clusters with different densities can be extracted in this way. Experimental results show that DENCH outperforms many popular algorithms, including K-means, CAST, and some model-based clustering algorithms.