Research On Prediction Method Of Drug Target Interaction Based On Heterogeneous Network

Identifying drug-target interactions is an important part of the modern new drug development process and a hot issue in medical research.Effective analysis of drug-target relationships plays an important role in drug repositioning,drug discovery,side effect prediction and drug resistance.In recent years,a lot of research has been conducted on drugtarget correlations,but predicting drug-target interactions in biological experiments is still tedious and time-consuming.Therefore,with the advancement of technology,predictions about drug-target relationships have become more and more popular among researchers.With the rapid development of machine learning and deep learning technologies,the use of efficient computational methods to predict drug-target interactions can effectively shorten the drug development cycle,reduce the blindness of new drug development,and provide theoretical guidance for biochemical experiments.Identifying effective DTIs is of great importance for the improvement of human medical technology.In this paper,we propose a K-Nearest Neighbor(KNN)model based on hybrid features and an aggregator model based on graph convolutional neural network for prediction of drug target interactions based on heterogeneous networks,respectively.According to the representation of drug compound information and protein sequence information on biological data,they cannot be directly input into the classifier as feature vectors.Based on this problem,we propose a KNN model based on hybrid features.The model uses a nonlinear similarity fusion method to combine different similarities with existing DTI networks and constructs a heterogeneous drug-target interaction network,while extracting feature vectors from the heterogeneous network to input into the KNN model for predicting DTI.finally,a quantitative description of the drug-target data is achieved so that the nodes maintain their intrinsic biological properties in the network.For large-scale,highly redundant data,traditional machine learning models cannot obtain good classification results,and most of the prediction studies on DTI in recent years have focused on drug nodes or target nodes,thus ignoring the drug-target relationships in heterogeneous networks.Therefore,a new method is proposed to predict drug-target interactions based on Efficient Graph Convolutions(EGC).A heterogeneous network is first constructed to extract the features of drug-target pairs,then a line graph is used to model the drug-target interaction relationship,and finally a graph convolution neural network is introduced to fuse multiple aggregators to predict potential drug-target interactions in the heterogeneous network.The model is able to learn multiple data features,reduce the dimensionality of the data and remove redundant information from the data.It is able to match well with hardware systems while maintaining accuracy,effectively increasing the speed of prediction.Compared with other DTI prediction algorithms,the proposed algorithm has significant advantages in speed and performance,further validating the reliability and efficiency of this paper’s algorithm in predicting drug-target interactions.