Drug–target interaction prediction: databases, web servers and computational models

Identification of drug-target interactions is an important process in drug discovery. Although high-throughput screening and other biological assays are becoming available, experimental methods for drug-target interaction identification remain to be extremely costly, time-consuming and challenging even nowadays. Therefore, various computational models have been developed to predict potential drug-target associations on a large scale. In this review, databases and web servers involved in drug-target identification and drug discovery are summarized. In addition, we mainly introduced some state-of-the-art computational models for drug-target interactions prediction, including network-based method, machine learning-based method and so on. Specially, for the machine learning-based method, much attention was paid to supervised and semi-supervised models, which have essential difference in the adoption of negative samples. Although significant improvements for drug-target interaction prediction have been obtained by many effective computational models, both network-based and machine learning-based methods have their disadvantages, respectively. Furthermore, we discuss the future directions of the network-based drug discovery and network approach for personalized drug discovery based on personalized medicine, genome sequencing, tumor clone-based network and cancer hallmark-based network. Finally, we discussed the new evaluation validation framework and the formulation of drug-target interactions prediction problem by more realistic regression formulation based on quantitative bioactivity data.