Prediction Of Drug-related Targets Based On Deep Learning

Drugs usually exert their actions by targeting corresponding proteins.Therefore,revealing underlying drug-target relations plays an essential role in drug development.Computational prediction of drug–target interactions(DTIs)in recent years,has become crucial to drug discovery or relocation.Compared with expensive and time-consuming in vivo or biological tests,computational methods can effectively identify potential DTIs and greatly reduce the range of drug candidate proteins.Several studies have already tried to use these kinds of methods to predict DTIs to reduce the cost and time of drug discovery or repositioning.Therefore,how to identify reliable drug-related candidate proteins is a subject worthy of research.Three DTIs prediction methods based on deep learning are proposed in this thesis,including a model of predicting drug-target interactions based on graph convolution and variational autoencoder(VGDTI)and a method based on dual attention mechanism model,called AttDTIPred as well as DTIPred,which infers drug-target interactions based on random walk and convolutional neural network.First,VGDTI utilizes graph convolutional neural networks to learn the topological structure representation of drug and protein nodes in a heterogeneous network based on drug-protein-related interactions and obtains a predicted DTI probability through a predictor.Then,the variational autoencoder is used to learn the current feature distribution of the similarities between drugs and proteins.This distribution is regarded as the feature representation of a drugprotein pair by VGDTI,and the potential interaction score is predicted.The two prediction scores are weighted average to integrate different information as to the likelihood of a drug protein pair interacting.Experiments show that integrating multisource heterogeneous data related to drugs and proteins can help improve the performance of DTIs prediction.The case studies of clozapine,quetiapine,aripiprazole,amitriptyline,and asenapine further shows that new drug-protein interactions can be predicted by VGDTI.AttDTIPred is a DTI prediction method based on the dual attention mechanism that integrates multiple similarities,interactions,and associations related to drugs and proteins.It learns to obtain low-dimensional vector representations of drug and protein nodes based on an encoding and decoding framework composed of a multi-layer fully connected network and a feature-level attention mechanism to integrate two drug similarities,protein similarities or interactions,and DTIs,respectively.The reduced dimensionality of the drug and protein vector representation is represented as a feature sequence,which is full of feature information related to drugs and proteins.At the same time,a prediction model based on one-dimensional convolutional neural networks(1D-CNNs)is proposed to further capture the abstract information represented by the low-dimensional vector of drug and protein nodes.Then,the attention mechanism module is also proposed to take into account which subsequences in the low-dimensional sequence of drugs and proteins are more important for DTIs when predicting DTIs.These drug and protein feature vector representations after the 1D-CNNs module and the attention mechanism module are spliced together as the input of the fully connected network to obtain our final DTIs prediction score.The experimental results show that AttDTIPred obtains better prediction performance than other state-of-the-art methods for the prediction of drug-target protein interactions.In addition,case studies of clozapine,quetiapine,aripiprazole,amitriptyline,and asenapine further demonstrate that AttDTIPred has the ability of predicting potential drug-protein interactions.A DTI prediction method based on random walk and convolutional neural network was proposed and referred to as DTIPred.DTIPred not only takes advantage of various original features related to drugs and proteins,but also integrates the topological information of heterogeneous networks.The prediction model is composed of two sides and learns the deep feature representation of a drug–protein pair.On the left side,random walk with restart is applied to learn the topological vectors of drug and protein nodes.The topological representation is further learned by the constructed deep learning frame based on convolutional neural network.The right side of the model focuses on integrating multiple original similarities and interactions of drugs and proteins to learn the original representation of the drug–protein pair.The results of cross-validation experiments demonstrate that DTIPred achieves better prediction performance than several state-ofthe-art methods.During the validation process,DTIPred can retrieve more actual drug–protein interactions within the top part of the predicted results,which may be more helpful to biologists.In addition,case studies on five drugs further demonstrate the ability of DTIPred to discover potential drug–protein interactions.