Study On Drug-target Interactions Prediction Based On Multimodal Deep Autoencoder

Drug targets are biological macromolecules in the human body that are related to diseases,and can be acted upon by drugs enabling drugs to play their medicinal functions.Drugs achieve the therapeutic effect of diseases by combining with the targets of specific diseases.The selection and identification of drug targets is the first step of drug development,and the discovery of drug targets helps people further understand the mechanism of drug operation,drug side-effects and disease pathology.The prediction of drug-target interactions(DTIs)is the main way to discover drug targets.Therefore,the prediction of drug-target interactions has important theoretical value and application significance in drug development and disease treatment field.The existing deep learning-based DTIs prediction methods only consider the similarity of chemical structure feature between drugs and the similarity of amino acid sequence feature between targets,ignoring the impact of various similarities on the prediction results.Studies have shown that the similarity of gene expression induced by drugs,the similarity of the drugs side-effects and other similarities,are directly related to drug-targets interactions.These similarities provide valuable multi-omics information for the DTIs prediction.Moreover,capturing global structure of similarity network is helpful to extract manifold information between network nodes,which can improve the prediction accuracy of methods.To solve above problems,this paper integrates a variety of similarities of drugs and targets,and constructs a drug-target prediction model based on multimodal deep autoencoder.The research contents are as follows:(1)In order to solve the problem that deep learning-based methods ignored multiple similarities of drugs and targets,a drug-target interaction prediction model based on multimodal deep autoencoder(MDADTI)is proposed.Firstly,multimodal deep autoencoder was used to fuse multiple similarities of drugs and targets,including chemical structure similarity,drug-induced gene expression similarity of drugs,amino acid sequence similarity and functional annotation similarity of targets and so on,to learn the deep-level features of drugs and targets.Then DNN was used to predict drug-target interactions.Theexperimental results show that MDADTI model can effectively utilize the complementary information in multiple similarities and improve the accuracy of DTIs prediction.(2)In order to solve the problem that the drug-target interactions prediction model based on multimodal deep autoencoder ignores the global structure information of similarity networks while predicting DTIs,a drug-target interactions prediction model based on global structure is proposed.First,the random walk with restart method is used to calculate the topological structure features of drugs and targets for each similarity network.Then,the positive mutual information method is used to calculate the similarity between topological features,and the topological similarity matrices of drugs and targets are obtained to capture the global structure information of similarity network.Finally,the MDADTI model is used to fuse multiple topological similarity matrices of drugs and targets and predict drug-target interactions.The experimental results show that the model can further improve the accuracy of DTIs prediction,which has certain reference significance for the related research of drug development.This model makes full use of the complementary information and global structure information of multiple similarities and realizes the automatic learning of drug-target features.It can effectively predict unknown drug-target interactions,and provide a novel idea for studies related to drug development and disease treatment.