Skip to main content
. 2010 Jun 30;20(2):026103. doi: 10.1063/1.3455188

Figure 2.

Figure 2

Comparison of cluster distributions and yield of biological information by REMc, Hc, and KMc. (a) Cluster size distributions from each of seven different clustering methods. With the cluster number fixed at 17, Hc results in a wider range of cluster sizes relative to other methods. (b) The output from 15 runs of CLUSTERJUDGE (CJ) using the entire result of each indicated clustering method as an input. 17 clusters, the number predicted by REMc, were assumed for each method. The p-value refers to t-test results comparing distributions of CJ scores between REMc and each other method. Clustering method abbreviations are REMc (recursive expectation-maximization), KMc_Euc (K-means with Euclidean distance), KMc_Pc (K-means with Pc), Hc_Pc_comp (hierarchical with Pc and complete linkage), Hc_Euc_comp (hierarchical with Euclidian distance and complete linkage), Hc_Pc_avg (hierarchical with Euclidian distance and average linkage), and Hc_Euc_avg (hierarchical with Pc and average linkage).