Fig. 1.

The protein complex detection framework. (a) Disease causing proteins and PPI data are retrieved from the literature. (b) Network propagation is executed starting at the causal proteins of a certain disease, yielding a ranked list of proteins which is then filtered using an empirical-based P-value. (c) Likelihood scores are assigned to all network PPIs, designed to prefer interactions that are more likely to appear in a protein complex model compared to a random model. The input for the detection phase is the weighted subnetwork induced by the proteins that passed the filtering. (d) High scoring clusters are detected using an integer linear programming algorithm