Research On Prediction Of Drug-Target Interaction And Binding Affinity

Drug-target interactions and their binding affinity are important indicators for drug research and development.Effective drug-target interaction data provide strong data support for the identification of toxic and side effects in drugs discovery,drug relocation and precision medicine.Traditional biochemical experiments are insufficient to meet the current demand for drug research and development in terms of time,economy,and data scale.The application of computational technology in this field has also emerged.The post-genomic era has produced massive amounts of biological data,which provide a research basis for predicting the association between drugs and targets,and computer technology provides effective auxiliary tools.Computer-aided prediction of the correlation between drug targets has the advantages of fast calculation process,accurate calculation and full automation.This method resolves the prediction problem of the correlation between drugs and targets into a process that can be calculated and visualized,to a certain extent,reducing the blindness in the drug screening process,thereby shortening drug development time and improving drug development quality.In this dissertation,two aspects of drug-target interaction relationship and drug-target binding affinity were studied,based on the computational technology.The research contents are as follows:1.An convolutional neural network is proposed to predict drug-target interactions.The interaction between drug and target is the key step in drug screening.In this work,a prediction model was constructed based on convolutional neural network which drug molecules were characterized by PaDEL-Descriptor software,and target proteins were characterized by amino acid coding and Moran autocorrelation coefficient.We used two datasets to train the model and the experimental results showed that the model had good prediction performance in multiple perspectives.The third dataset was used to test the model,and the results showed that the proposed model has higher prediction accuracy than other deep learning models.The model was proved to be able to identify the drug-target interaction under various conditions.2.An algorithm of integrating machine learning with network fusion was proposed to the prediction of drug-target binding affinity.The pattern and state of drug-target interaction can be further explored by analyzing drug-target combination.In this work,three datasets were used to calculate the multi-angle information of drugs(targets),and then network similarity fusion was developed to remove the redundant information,so as to realize the complementarity between different feature information.The resulting features were training with XGBoost to build a model that can predict the binding affinity between drugs and targets.The results of this study confirmed that the model achieves higher prediction accuracy than other methods.By integrating multiple similarity information,it was found that different similarity information can be complemented each other and the integration of these information can achieve high prediction effect.In summary,this dissertation studies the drug-target interaction relationship and their binding affinity.Based on machine learning and deep learning methods,two prediction models were constructed respectively,and good prediction performance was obtained.The purpose of this study is to provide research ideas for the correlation between drugs and targets,and to provide a data processing model for drug screening,so as to reduce the cost and loss in the process and improve the efficiency of model.