We present a method to discover signaling pathways, quantify the relationship of preselected source/target nodes, and extract relevant subgraphs in large scale biological networks. This is demonstrated over the hepatocyte growth factor (HGF) stimulated cell migration and proliferation in a keratinocyte-fibroblast co-culture. The algorithm (MCWalk) is implemented with random walks using Monte Carlo simulations. We extract a master network by overlaying case specific microarray data from the NCI Pathway Interaction Database (PID) using a fully automatic pipeline without any manual network construction, and uncover the association of HGF receptor c-Met nodes, differentially expressed (DE) protein nodes and cellular states. We show that the network has a scale-free structure and identify key regulator nodes based on their random walk traversal frequency. This property is shown to be very weakly correlated to node degree, contrary to what is expected from similar centrality measures. The differences with standard methods, such as shortest-path, commonly used in the analysis of such networks are discussed and compared with this approach, highlighting important pathways which are exclusively obtained with our random walks algorithm.
Category: Quantitative Biology