Background: Predicting interactions between drugs and target proteins is a key task in drug discovery. Although the method of validation via wet-lab experiments has become available, experimental methods for drug-target interactions (DTIs) identification remain either time consuming or heavily dependent on domain expertise. Therefore, various computational models have been proposed to predict possible interactions between drugs and target proteins. Usually, we construct a heterogeneous network with drugs and target proteins to calculate the relationship between them. However, most calculation methods do not consider the topological structure of the relationship between drugs and target proteins. Fortunately, Network Embedding Learning provides new and powerful graph analytical approaches for predicting drug-target interaction, which is considering both content and topology of network.
Results: In this article, we propose a relational topology-based heterogeneous network embedding method to predict DITs, abbreviated as RTHNE_DTI. We use the ideas of word embeddings to turn heterogeneous network with drugs and target proteins into dense, low-dimensional real-valued vectors. Furthermore, according to two different topological structure of the relationship between the nodes, we represent them separately by training two different models. Then the meaningful vectors represented for drugs and target proteins can be used to calculate the interaction of them easily. Results show that by considering topological structure and different relationship type of drugs and target proteins, RTHNE_DTI outperforms other state-of-the-art methods on both labeled network and unlabeled network.
Conclusions: This work proposes heterogeneous network representation learning for DITs prediction. To the best of our knowledge, this study first introduces relation classification to heterogeneous network embedding to improve predicting DTIs efficiently.