Abstract—Drug Target Interactions (DTIs) prediction research is one of the key links in drug development, and is of great significance to the fields of new drug development and drug relocation. In recent years, network representation learning technology has developed rapidly. Network representation learning is also known as Graph Embedding, which can be used in applications such as node classification, link prediction, and community discovery. The key challenge in this field is to effectively vectorize complex homogeneous or heterogeneous networks, and how to fully reflect the network structure, node relationships and connection information. Deep learning has made great achievements in the fields of speech recognition, computer vision, and natural language processing. There are also many research results in the field of DTIs prediction, but the combination of the two is relatively small, and it is worthy of in-depth study. This paper proposes DTIs prediction based on NV-DNN method. The drug-drug and target-target relationships are used to form the Jaccard matrix. The Node2Vec method is used to learn the feature vector representation, and then input it into the DNN network for deep learning to improve classification and prediction capabilities. Experiments show that AUC 0.89 is better than other common methods that only use the network structure information of drugs and targets, and do not use the attribute information. Due to the simple input of the model, it can predict drugs and targets with unknown attributes, so it has a wide range of adaptability.
 

Index Terms—drug-target interaction prediction, network representation learning, Node2Vec, DNN