An example of latent graphs used in the proposed EGM-SMCC model. (a) PPI latent graph, i,e., the original interaction network, where the ground-truth label of the center node v1 is "+", but it is difficult to predict the label (+ or -) for node v1 since it has the same number of positive and negative neighboring nodes; (b) even-step random walk latent graph, the directed neighbors with the same label ("+") to v1 have triangle edges (red lines), hence they are reachable from v1 using even-step random walks. On the other hand, the indirected neighbors (from the + nodes) in network are linked by creating edges (dash line) using even-step random walks; (c) prediction similarity latent graph using kNN graph construction. In the kNN graph, a node pair share an indirected edge if the two nodes are k-nearest neighbors. In the example, we set k = 3.