Automatic Context-Specific Subnetwork Discovery from Large Interaction Networks
Let three probes (i, j, and k) constitute the representative probeset of an MIS. We plot all the samples with the expression values of these probes in separate dimensions, and then we cluster the samples. If the samples in a cluster are mostly of one kind (such as or ), we can say that the cluster’s expression pattern represents the corresponding class (positive or negative). The ranking of an MIS producing such clusters should be high.