Finding effective subnetwork markers for cancer by passing messages

It is generally difficult to predict cancer outcome based on individual genes, and recent research results have shown that the use of pathway or subnetwork markers can improve the accuracy and reliability of such prediction. In this work, we propose a novel method for identifying subnetwork markers that can accurately predict cancer metastasis. The proposed method takes an efficient message passing approach to search for non-overlapping subnetwork markers in the human protein interaction network. Experimental results show that this method can identify robust subnetwork markers that may lead to enhanced cancer classifiers.