Agglomerative hierarchical clustering according to similarity in gene expression patterns in 35 clinical samples and 7 representative cell lines harboring BCR/ABL, TEL/AML1, or an MLL abnormality. For each cell line and clinical sample, total RNA and reference total RNA (Universal Human Reference; Stratagene) were treated with DNase (DNA-free; Ambion, Austin, TX), linearly amplified, fluorescently labeled (with Cy5- or Cy3-dUTP) and comparatively hybridized to an array containing approximately 43 000 features representing approximately 31 000 unique UniGene clusters. This data set was obtained by using 2 distinct array batches that were produced using 2 different batches of PCR amplifications of the clone inserts. The data for each gene was mean centered by batch. A list of arrays used and the batches to which they belong is contained in the Supplemental Materials. We replaced the gene expression levels obtained from duplicate arrays by their mean expression level. (A) The variation in expression is displayed as a variation in color¹⁰  for the 844 clones (representing 758 genes) for which the log₂ intensity ratio differed by at least 2 from its mean on at least 2 arrays. The color scale extends from 0.18-fold to 5.7-fold of the mean (-2.5 to 2.5 in log₂ space) as indicated in the upper-right corner. Gray represents omitted data. The arms of the dendrogram are color-coded to indicate the chromosomal translocation associated with each branch: purple indicates MLL; orange, TEL/AML1; and blue, BCR/ABL. The gray bar indicates data obtained from cell lines. (B) A subset of genes from panel A that are highly expressed in cell lines. Gene names are provided for selected named clones. (C) Clustering of the 35 clinical samples shown in panel A. The variation in expression is displayed for the 272 clones (representing 255 genes) for which the log₂ intensity ratio differed by at least 2 from its mean on at least 3 arrays. The broad features of the clustering patterns were robust to variations in gene selection criteria.