Drug-target interaction prediction with deep learning

OBJECTIVE The identification of drugtarget interaction(DTI) is an important process for drug discovery. It has been a hot research topic in the field of drug design to develop different computational models to predict potential DTIs and provide powerful complementary tools for biological experiments. METHODS Artificial intelligence has been applied to predict DTIs, however,there are still many obstacles that need to be overcome,such as low prediction accuracy, unreasonable selection of negative samples, and poor data reliability. In this paper, we developed a deep learning based method named DLDTIs to predict DTIs. By introducing high dimensional molecular fingerprints and protein descriptors, and then applying a probability matrix decomposition algorithm to generate negative sample data sets, we constructed a promising DTI classification model. RESULTS DLDTIs was comparable or superior to other methods against the test sets by achieving more than 90% of accuracy, specificity, sensitivity and area under the curve, respectively. CONCLUSION The probability matrix decomposition algorithm was beneficial for generating negative DTIs. DLDTIs may be served as a powerful tool for future drug target prediction.