Robust resolution of fuzzy clusters. (A) Fuzzy cluster network showing the AutoSOME transcriptome analysis of 11 Acute Myelogenous Leukemia cell lines (AML, circles) and 27 Acute Lymphoblastic Leukemia cell lines (ALL), comprised of 19 B-cell lines (ALL-B, diamonds) and 8 T-cell lines (ALL-T, rounded-squares) [9, 24]. Nodes are colored by cluster membership, and numbers represent individual cell lines as originally ordered . Cytoscape network parameters are the same as those used in Figure 4. (B) The same AutoSOME output is shown by a hierarchically reordered heat map, illustrating the considerable fuzziness of cell-cell co-clustering over 500 ensemble iterations (E = 500). The diagonal represents self-comparison of each cell line while off-diagonal colors denote cell-cell pairwise affinities as defined in Figure 4. (C) A hierarchically reordered heat map of final cluster assignments averaged over five independent AutoSOME runs (each E = 500) illustrates the consistent resolution of the fuzzy clusters into discrete, non-overlapping clusters. Hierarchically reordered heat maps shown in Panels B and C were generated using PermutMatrix software .