Cluster analysis of AK-AML samples using the CvN method and the CvC method. (A-B) Example of PCA plots of individual gene expression profiles of complex karyotype AML (A) and t(15;17) AML patient samples (B) projected to the gene expression-based map of the normal hematopoietic hierarchy, using the CvN method (see “Bioinformatics analyses” and Figure 1). Only the 2 first (ie, PC1 and PC2) PCs are given in the PCA plot. A line indicates the nearest normal counterpart for each of the AK-AML samples. (C-D) Unsupervised clustering of AK-AML. PCA of AK-AML based on genes identified by the CvC- (C) and CvN method (D). Genes were selected by variance (1545 and 1449 probe sets in (C) and (D); respectively). ANOVA analysis of the segregation of the clusters using the first 5 PCs reports P values of .49 for the CvC method (inter-group variance: 6.08, intra-group variance: 495.85) and .004 for the CvN method (inter-group variance: 115.58, intra-group variance: 1685.95) for the CvC and CvN methods, respectively. (E) ROC curves (classification performances) for 2 published and 1 novel AML t(11q23) gene signatures. Areas under ROC curve (AUCs) are reported in the graph. (F). Heat map representing the degree of enrichment (-log10[P value]) in the 1% upregulated genes in AML patients⁹  of known and novel AK-AML gene signatures.