Figure 3

The gene-expression profiles of cell lines compared to normal and tumor tissues. (a) Projection of each sample in dataset I into SVD space drawn by the correlation of each sample to SVD eigenarray 1 (x-axis) and 2 (y-axis). The normal tissue samples of CNS origin from two laboratories (green squares, Hsiao et al [17]; black squares, Ramaswamy et al. [16]) were overlapping, as well as the tumor tissue samples (red squares, Ramaswamy et al. [16]). The cell lines were separated from tissue samples by the first SVD eigenarray. Samples of lymphoma and leukemia origin were also separated in the SVD analysis. (b) Projection of each sample in dataset II into the SVD space drawn by the correlation of each sample to SVD eigenarray 1 (x-axis) and 2 (y-axis). The cell lines (crosses) were separated from tissue samples. Whole blood samples were distinctly clustered close to the cell lines. (c) Other separation of normal samples. Significance analysis of microarrays (SAM) was used to identify differentially expressed genes between cell line and tissue samples in dataset I. The number of statistically significant genes (x-axis) as a function of the median and 90th percentile of the FDR (y-axis) estimated based on 1,000 permutations. (d) SAM analysis of cell line versus tissue samples in dataset II. Identical parameters as in (c). (e) Plot of the degree of differential expression between cell lines and tissues for each gene in dataset I (x-axis) versus dataset II (y-axis) respectively. The degree of differential expression was measured using the signal-to-noise metric [23].