Figure 1
From: A network approach for low dimensional signatures from high throughput data
(a) A simple 2D ideal model in which single-feature classification performance fails in predicting higher-dimension classification performance. Both features (gene expression 1 and gene expression 2) badly classify in 1D but have a very good performance in 2D. Moreover, the classification can be easily interpreted in terms of combined higher/lower expression of both probes. (b) Activity of a biological feature (e.g. a gene) as a function of its expression level: top—monotonically increasing, often also dichotomized to an on/off state; bottom—“windowed” behavior, in which the two activity states do not depend monotonically on expression levels. X axis: expression level, Y axis: biological state (arbitrary scales).
