High Accuracy Drug-Target Protein Interaction Prediction Method based on DBN

Drugs may have multiple drug targets, and the most of targets are composed of different proteins. Therefore, the study of drug-target interaction (DTI) prediction has important meaning in drug repositioning, drug development time shortening and the cost of drug research and development reducing. Most of the existing methods are based on shallow learning model. The prediction accuracy is not high. In this paper, we proposed a deep belief network-based DTI prediction algorithm: we extracted extended connected fingerprint of the drug from the molecular structure. And then, we extracted the structure characteristics of the three peptide of the protein from the amino acid sequence of the protein. At last, we train the deep belief network by the characteristic vector extracted from drugs and proteins. In our proposed method, we fully use of the characteristics in the deep learning network and integrate the empirical feature selection into the deep belief network. Base on the public data set and compared with the state-of-the-art approaches, the experimental results show that our method outperforms the other algorithms in massive data sets.