Experimental data from protein microarrays or other targeted assays are often analyzed using network-based visualization and modeling approaches. Reference networks, such as a graph of known protein-protein interactions, can be used to place experimental data in the context of biological pathways, making the results more interpretable. The first step in network-based visualization and modeling involves mapping the measured experimental endpoints to network nodes, but in targeted assays many network nodes have no corresponding measured endpoints. This leads to a novel problem - given full network structure and a subset of vertices that correspond to measured protein endpoints, infer connectivity between those vertices. We solve the problem by defining a semi-lazy random walk in directed graphs, and quantifying the mean first-passage time for graph nodes. Using simulated and real networks and data, we show that the graph connectivity structure inferred by the proposed method has higher agreement with underlying biology than two alternative strategies.
