Drug-Target Interactions Prediction Based On Deep Learning

In modern pharmacological research,identification of effective drug-target interactions is one of the crucial steps in the process of new drug development and drug repositioning.It is important for the analysis of drug molecules,the understanding of drug mechanism of action,and the accuracy and safety of drug efficacy.However,validation of feasible drug-target interactions based on traditional biochemical assays is a time-consuming,costly and unsuccessful process.With the advances in biology and science,the availability of large-scale genomic,pharmacological and chemical data has provided new directions for drug-target interaction prediction.How to design computation-based methods to predict drug-target interactions has become a hot research topic in the field of bioinformatics.Compared with traditional biological experimental methods,computational-based methods have the advantages of saving human and financial resources and improving prediction efficiency.In this paper,we investigate the prediction of drug-target interactions based on the biological and chemical data of drugs and targets,combined with relevant deep learning techniques:(1)In this paper,we propose a drug-target interaction prediction method MHSADTI based on multi-head self-attention and graph attention network,which captures the topological structure information of drug molecular graphs and effectively solves the influence of noisy connections on the extraction of node information in the graph.MHSADTI is based on the multi-head self-attentive module to extract amino acid sequence information of targets,and more effectively captures the contextual association information present in the sequences.Experimental results on multiple datasets show that the MHSADTI outperforms many advanced prediction algorithms in terms of prediction accuracy.Also,visualization of the intermediate results learned by the model indicates that MHSADTI has learned biological level knowledge,which is important for guiding new drug design.(2)In this paper,we propose a prediction method based on the fusion of drug-target independent features and interactive features,IIFDTI.The IIFDTI mainly consists of convolutional neural network,graph attention network and bidirectional encoder-decoder structure,in which convolutional neural network and graph attention network are used to extract independent features of drug and target respectively,and bidirectional encoderdecoder structure extracts interactive features of drug-target pairs.IIFDTI finally fuses the four feature vectors and inputs them to the fully connected layer for drug-target interaction prediction.Experimental results on multiple datasets show that the IIFDTI achieves better results than various other advanced drug-target interaction prediction methods on several metrics.By visualizing the attentional mechanisms in IIFDTI,the results show that the model has learned the biological perspective and validated the practical implications of the model by performing denove experiments on new drugs and new targets.