Prediction of Protein Function using Graph Container and Message Passing

We introduce a novel parameter called container flux, which is used to measure the information sharing capacity between two distinct nodes in a graph. We also formulate a new equation for protein function prediction by integrating the container flux as an information sharing component. Based on the scale free characteristic of protein interaction network, we propose that these proteins of high degrees most likely be the exemplars for difference clusters. By further exploration, we reveal an interesting connection between the global optimization of our prediction equation and the exemplar-guided clustering problems. Our preliminary experimental results support our methods