We tested the stability of the dendrogram structure while changing the number of genes selected for calculating the dendrogram. We selected N in the range 10-44. For each N we repeated the procedure above and stored the selected genes and dendrogram structure. We then examined all branches (junctions) of the reference tree (N = 28) and compared them to the corresponding branch in the test tree. We derived two stability criteria (1) branches with leafs below having 90% overlap in test compared to reference, (2) branches with exactly the same set of clusters and the same order of the leafs. For each branch we calculated the fraction of cases that either criteria (1) or (2) occurred. More than 65% of the 264 branches had probability of 1 and 94% had probability greater than 0.5 based on criterion (1). Based on the more stringent criterion (2) more than 50% had probability of 1 and about 85% greater than 0.5.
