All protein targets of a compound might not be identified during the compound development stage. The expected side effects of compounds while using them in treatments might be observed due to the binding of compounds to off-target proteins and the biological processes triggered by these off-targets. If the protein targets of compounds would be identified more comprehensively, the side effects observed after a disease treatment might be also reduced. The aim of this study is to identify potential targets of a compound with a computational method. The proposed method will compute potential off-targets of a compound by using gene expression data of compound-treated cells on protein-protein interaction networks. This method mimics the cellular processes in terms of topological relations by integrating protein interactions and the transcriptome data of a given compound. The method first maps simplified compound effects on the network, and then computes various network centrality metrics to suggest the most probable targets of a new compound. The experiments revealed that the type of interaction network dramatically effects the target identification performance of the method. Furthermore, network centrality metrics might produce variable results based on selected confidence cut-offs and the network type. The proposed method simply implements graph-based data mining on integrated biological resources to reduce time and cost of computer-based compound development.