Unsupervised clustering analysis. A) Phylogenetic plot of the samples using all genes in the database (17,589 in total). B) Similar plot based on 3518 genes filtered by interquartile range (IQR > 0.9) and coefficient of variation (CV > 0.1) to remove genes with small intensity variations among the studied samples. C) Consensus plot based on the filtered genes; 500 clustering trees were generated by bootstrapped genes. The numbers on the branching lines are the percentage of times that the specific cluster occurred among the 500 trees. Normal samples from either renal pelvis or bladder always clustered with their respective groups (100%), whereas upper- and lower-tract UCs demonstrated frequent overlap. D) Heat map of SAM-selected genes that are the most differentiated between UC samples in the C3 and C4 groups identified in the unsupervised study (Figure 2A and B).
Zhang et al. BMC Medical Genomics 2010 3:58 doi:10.1186/1755-8794-3-58