Figure 4

Comparison of cluster algorithms using the Rand measure. We estimated a pairwise degree of similarity between two clustering results both with identical cluster numbers produced by two different algorithms using the adjusted Rand index. The three heatmaps shown in this figure each visualize the mean of all Rand indexes computed for cluster numbers from two to 50 for experiment A and C. For experiment B, we selected a different highest cluster number--here 100 was chosen--in respect to the experiment's dataset size and, hence, its higher number of quantified proteins. In all three experiments, notably, a high degree of similarity is observed between the three cluster methods K-means, Neuralgas, and HCA using Ward's Linkage; in experiment C, in addition, to fuzzy C-means, Complete- and less pronounced to Average-Linkage (the two latter with Euclidean distances). In experiments A and B, a slight similarity is, furthermore, found between Single- and Average-Linkage (likewise with Euclidean distances).