Experimental and clinical descriptors nevertheless were provided for all chip data files in addition to digitally archived microscopy images of histological preparations. Prior to carrying out discovery research using these data, rigorous quality control testing was applied to these data. A total of 454 microarrays met quality control requirements and were judged suitable for this discovery research. Further details of the quality control procedures applied to these data re provided in Supporting Information S1 Quality Control Analysis. Description of the tissue phenotypes for these discovery data is shown in Table 3 with cancer phenotype breakdown in Table 4. Table 3 Phenotypic breakdown of clinical specimens used in this study. Table 4 A description of tumor specimens by stage for discovery tissues.

Phenotype-specific expression patterns In addition to standard differential expression analysis, we introduced an analytical technique designed to filter differentially expressed probeset candidates for transcripts that we hypothesized were qualitatively ��turned-on�� in one phenotype class and qualitatively ��turned-off�� in a comparator phenotype. For this method, identification of ��off�� genes was based on the relatively simple assumption that most genes in a given tissue were not constitutively expressed above a nominal relatively low background level. Consequently, a microarray designed to hybridize to the full human transcriptome should therefore not exhibit transcript-specific binding for most probesets in a given experiment.

Conversely, the fluorescent intensity of probesets that hybridized to the balance of non-expressed transcripts should reflect ��non-specific�� probeset-transcript hybridization. The assumption that a large fraction of the probesets for any given experiment were not transcript-specific signals provided means to estimate a theoretical on/off threshold for genes in full-genome experiments such as used here for discovery. The mean expression level for all 44,928 probesets in the 454 discovery microarrays were ranked and the probeset value corresponding to the 30th percentile value across the data was chosen as the threshold for transcriptional silence. This threshold represents a conservative upper-bound estimate of non-specific or background expression. Validation Data: Tissue specimens For all validation (i.e.

hypothesis testing) experiments, independently collected fresh frozen tissue specimens were obtained from a tertiary referral Batimastat hospital tissue bank (Flinders Medical Centre, Adelaide, SA Australia). A description of cases used for validation testing is shown in Table 3. This study was approved by the Research and Ethics Committee of the Repatriation General Hospital and the Ethics Committee of Flinders Medical Centre. Written informed patient consent was received for each tissue studied.