Figure 2

Illustration of the set-cover based algorithm for the identification of one-vs-all discriminative subnetworks. Here, the coverage provided by a hypothetical three-gene subnetwork is shown. In the left panel, the distribution of the expression levels of each gene across all samples is shown. We first compute the mean (μ _i ) and the standard deviation (σ _i ) of this distribution for each gene g _i . Subsequently, we identify samples that are positively or negatively covered by each gene. A gene g _i with expression greater than μ _i + α * σ _i in a sample is said to positively cover that sample, while a gene g _i with expression less than μ _i - α * σ _i in a sample is said to negatively cover that sample (we set α = 2 in our experiments). The negative and positive cover sets for each gene and the subnetwork composed of g₁, g₂ and g₃ are shown on the right. In this example, this subnetwork negatively covers (all samples in) Stage B.