Figure 4

Aspects of gene expression patterns. Top ranked genes from four different methods are shown (A: N-MEG, B: multiple regression, C: two-way ANOVA, D: Suzuki set, E: Ellsworth set, and G: Abba set). Genes in the first row are up-regulated genes in lymph node positive samples in each study, the second row are down regulated genes (except in B; ANOVA does not give directional results). Lymph node progression is denoted in X-axis, and the relative expression values against normal breast samples are denoted in Y-axis (in log 2 based fold change). Four different colors are used to discriminate different tumor sizes (T stage 1 to 4). Expression patterns of two genes from N-MEG and multiple regression (FLJ32549, AKR1D1, TMEM65 and MGC35361) show consistent increase (top row) or decrease (bottom row) in all tumor sizes, while top-ranked genes from other studies (D~F) show either inconsistency among different tumor sizes (D up, E and F up) or disagreement over phenotypic traits (D down, E down, and F up). We can see that the multiple regression model find more linear relationships (B) and ANOVA set contains significant fluctuations (C).