Figure 6

Estimation of optimal cluster number by gap-stat analysis. The gap statistic for the hierarchical clustering of the genes analyzed by RT-PCR. The within-cluster dispersion (an error measure) decreases monotonically as the number of clusters increases. If the data do tend to form a certain number of clusters, say k, then the decrease in the within-cluster dispersion will flatten markedly thereafter. The gap statistic calculates the value(s) of k at which this happens in order to estimate the optimal number of clusters [15]. In the RT-PCR data, two well separated clusters can be identified because of the local maximum at k = 2. However, the gap function rises again with a strong increase at k = 4 and k = 6, suggesting the existence of less well separated clusters within the main clusters.