GEPs of MDS samples, de novo AML, and “none-of-the-targets” cases. (A) In the PCA, 406 cases are displayed based on the 534-probe set signature from the DC model. The first 3 principal components accounted for 28.8% of variation of the data (component 1 = 16.9%; component 2 = 6.6%; component 3 = 5.3%). Each sphere represents a single GEP. The AML (n = 202) and “none-of-the-targets” (n = 69) samples are colored according to the initial diagnosis. The shape of the AML ellipsoid was determined by the variability within the AML samples, and the ellipsoid was drawn to surround the samples within 2-fold SD. In the MDS group (n = 135), cases called by the diagnostic classifier as AML (n = 31) are further distinguished. Detailed information on the classifier probe sets is available online. (B) The agglomerative hierarchical clustering yields an entire hierarchy of clusters for all samples in the dataset. Euclidean distance was used to measure the dissimilarity between AML (n = 202), MDS (n = 135), and “none-of-the-targets” (n = 69) samples. Ward's minimum-variance method was used to determine the hierarchy and to define the groups. The average width of the clustering structure was represented as dendrogram in the clustering tree. The samples are annotated according to diagnostic category (Class), laboratory where the microarray analyses were performed (Center), and results of the DC model (Classification).