Fig. 1.

Workflow of the proposed method. 1) We utilized PPI data from the HIPPIE database and DDI and DTI data from the DrugBank database. 2) From both the interactome and DTI data, we constructed a heterogeneous network. 3) We then conducted a random walk with restart algorithm for all nodes (drug and target nodes on each interactome) and weighted features of nodes with the result of the RWR. Weighted features are getting similar on feature spaces for nodes which are closely connected in a network. 4) From the weighted features, we generated positive DTI pair vectors from a bipartite DTI graph and random negative DTI pair vectors. We trained the cubic kNN with the positive and negative DTI pairs