Deep Belief Network-based Drug-target Interaction Prediction

Protein is a significant material basis of life activities and plays an important role in all life activities.Drug-protein interactions(DTI)are common ways for drugs to treat diseases.Drug targets are binding sites for drug and biological macromolecules.According to polypharmacology,drugs may have multiple drug targets,and most of targets are composed of proteins.Therefore,the study of DTI prediction is conducive to drug repositioning and shortening drug development time.Experimental validation is a traditional method for studying drug-target interaction,which is a costly and time-consuming way in drug discovery.Thanks to DTI prediction base on compute method can reduce the cost of prediction,how to use computer simulation and pattern recognition on DTI prediction becomes an important work.A large number of DTI prediction base on compute method have emerged at present.This paper proposed a deep belief network-based drug-target interaction prediction algorithm.The network’s input is combination of the extend connectivity fingerprint of drug and the tripeptide composition of protein,and the network’s output is the probability of DTI.Finally,a threshold determines whether there is interaction between the drug and the protein.Our method can predict potential interactions of new drugs or new proteins.Our method was firstly compared with state-of-the-art DTI prediction algorithm on four standard datasets which was established by Yamanishi.Then the performance of the proposed method was verified on Drugbank dataset at experimental stage.The experimental results show that the method in this paper can improve accuracy of DTI prediction.At last,we build the DTI prediction software based on our algorithm which can show the performance of deep belief network-based DTI prediction and help researchers study drug-target interaction.