Box 1. Schematic overview of subnetwork identification

Protein–protein interaction networks are used to assign sets of genes to discrete subnetworks. Gene expression profiles of tissue samples drawn from each type of cancer (i.e., metastatic or non-metastatic) are transformed into a 'subnetwork activity matrix'. For a given subnetwork M_k in the interaction network, the activity is a combined z-score derived from the expression of its individual genes. After overlaying the expression vector of each gene on its corresponding protein in the interaction network, subnetworks with discriminative activities are found via a greedy search. Significant subnetworks are selected based on null distributions estimated from permuted subnetworks (see Materials and methods). Subnetworks are then used to identify disease genes, and the subnetwork activity matrix is also used to train a classifier.