Fig. 3.

Internal validity indices for combinations of workflows and clustering algorithms, for k=3–20. The values of all indices are scaled and re-centred around 0 for better visualization. Maximum values of Dunn2 and silhouette, and minimum values of McClain–Rao indicate optimal number of clusters. The bold black vertical lines indicate the consensus optimal number of clusters as indicated in Table 1. The silhouette index is not defined for k>13 in average- and complete-linkage clustering of the SNP2 workflow.