Figure 2

Semi-supervised extraction of significant genes from the same artificial data shown in Fig. 1(A). The true labels of the 16 samples are 1 for the earlier eight samples and 2 for the rest, but three samples of both classes are assumed to be unknown (labeled as * in the figure title). This experiment evaluates how many genes in the truly active 1,600 genes of group (4) can be detected by four significance scores: LR-S, SAM, ODP-S, and ODP-SS. Note that the genes in group (3) are regarded as inactive in this experiment because the high/low patterns are different from the true label pattern. Therefore, there are 4,800 inactive genes (groups (1), (2), and (3)) and 1,600 active ones (group (4)). (A) ROC curves of the four scores. The number for each score denotes the AUC (area under the curve) values. (B) Distributions of the ODP-SS scores for the four gene groups, (1), (2), (3), and (4). Horizontal axis is shown in log-scale.