Prediction Of Drug Target Interaction Based On Graph Convolution

Predicting the interaction between the drug and the target is very important for the development of new drugs,it can speed up the process of drug development and reduce the required cost.However,traditional clinical trials require a lot of manpower and material resources,which are expensive and time-consuming.Traditional machine learning methods mostly use one-dimensional strings or descriptors to represent drug compounds and protein targets.This manual feature-based method cannot completely extract the features of drug molecules and protein targets.Deep neural network does not consider and make full use of the internal structure information of the drug in the process of feature extraction.Therefore,it is urgent to study new methods to predict the interaction between the drug and the target.Graph neural network extends the neural network model to any structure diagram.A drug can be seen as a molecular diagram composed of atoms and chemical bonds.The graph convolutional neural network can automatically obtain the drug-related features from the drug molecular map through the convolution operation,and can make full use of the atomic information and internal structural information of the drug.The rise of graph neural networks provides a new solution for drug-target interaction prediction.Based on this,this paper proposes a drug-target interaction prediction method based on graph convolutional neural networks,which can capture the structural information of drug compounds and use the chemical background of protein sequences to solve the drug-target prediction problem.The main work of this paper is as follows:(1)A model GraphCPI based on graph convolutional neural networks and traditional convolutional neural networks for predicting drug-target interactions is proposed.This model uses graph convolutional neural networks to obtain the atomic and structural features of drugs,and uses traditional convolutions.The neural network obtains the characteristics of the target.In order to verify the effect of the Graph CPI model,this article uses the Human and C.elegans data sets,which are widely used in the pharmaceutical field,and compares them with traditional machine learning methods and the latest methods.The experimental results show that the model proposed in this paper has certain advantages.At the same time,this article carried out an ablation experiment.The ablation experiment made two variants of the Graph CPI model.The ablation experiment showed that each part of the model has an important effect on the model as a whole.(2)The GraphCPI model proposed in this paper can be used as a general model.Graph convolutional neural networks and traditional convolutional neural networks can be regarded as the building blocks of the model,allowing the use of other more powerful graph convolutional neural networks for replacement.In this article,we used GIN,GAT,and GCN for experiments.The experimental results verify that a variety of graph convolutional neural networks can be integrated into the GraphCPI model,showing the versatility and stability of the GraphCPI model.