Predicting Of Drug-drug Interactions And Synergistic Efficacy Based On Graph Neural Networks

In recent years,with the continuous improvement of national health awareness,people have begun to pay more attention to drug safety.In modern clinical medicine,most diseases are treated with drug combination therapy.However,due to the differences in the mechanism of action of the drugs,adverse drug side effects may occur,and even threaten the life safety of patients in severe cases.Therefore,in the development of new drugs,determining the interaction and synergistic efficacy between drugs is an integral part of the drug combination development step.However,drug discovery and development is a high-cost,high-risk and extremely time-consuming task,and traditional laboratory prediction methods have certain probabilistic uncertainties and blindness.Therefore,there is an urgent need to establish an efficient computational method to predict the interaction and synergistic effects of drug combinations.This thesis uses machine learning algorithms such as Graph Neural Networks(GNN)and attention mechanisms to predict drug-drug interactions and synergistic efficacy.The main research contents of this thesis are as follows:1.Most of the existing combination drug screening models only consider the similarity of drug structure and function and conventional signaling pathways,and do not fully consider information such as drug molecular networks.For the above problems,this study constructed a GCNDDI model for predicting drug-drug interactions based on graph convolutional network methods.The GCNDDI method represents the drug molecule as a graph,the atoms in the drug molecule are represented as graph nodes,and the chemical bonds between atoms are represented as graph edges.At the same time,in order to solve the problem that the features of graph nodes are not obvious in some current methods,based on the end-to-end learning of graph convolutional networks,this study introduces the R-radius subgraph method to extract the features of drug molecules.The drug interaction(DDI)data information in this thesis comes from the STITCH database.After data preprocessing,three sub-data sets,CCI900,CCI800,and CCI700,are constructed for model training.The prediction results of 100 times of five-fold cross-validation show that the AUC value of this method is above 0.93 on all three datasets,which is superior to other drug interaction prediction algorithms.At the same time,the results of the case study also showed that GCNDDI can predict drug interactions well.2.Although some methods for predicting the synergistic effect of drugs have been proposed,they usually directly use the chemical structure of the drug as a feature for prediction,without considering that the occurrence of synergistic effect between drugs is essentially due to the synergy between the molecular substructures of drug compounds caused by the effect.Therefore,on the basis of drug interaction,this study proposes a prediction model GATSyn based on the graph attention network method to construct the synergistic effect between drugs.The GATSyn method decomposes compound molecules into substructures,and predicts the synergistic effect between drug pairs by predicting the synergistic effect between the substructures.Meanwhile,in order to overcome the influence of noisy neighbor nodes,GATSyn applies graph attention network to assign different weights to nodes.In this thesis,the drug synergy prediction dataset is obtained from the Drug Bank database,and the model is trained on the constructed two sub-datasets DB1 and DB2.The experimental results show that the AUC values of the GATSyn method on both datasets are above 0.97,the performance is better than other prediction algorithms,and it is an effective tool for drug synergistic efficacy prediction.