Drug Clinical Decision Support System Based On GCN Fusion Multiple Association Network

Objective: It is an important and urgent task to find new therapeutic drugs or treatment schemes.The key of research is to determine the correlation between drugs and diseases.In order to solve this problem,researchers have used machine learning and deep learning methods to identify the relationship between drugs and diseases in recent years.However,most of the existing methods are to construct drug networks and diseases networks respectively,and predict the new drug disease relationship based on the known relationship between drugs and diseases,without considering the drug-target and drugprotein interaction relationship.Therefore,it is of great significance to develop an efficient and high-precision drug-disease association prediction method based on multiple association networks.Methods: In this study,a set of drug clinical decision support system is developed,which integrates the associated information of drugs,diseases and targets,and adopts the graph convolution neural network based on attention mechanism.Systematically train and construct the relevant information of drugs,diseases and targets from literature sources.Through screening the collected correlation information of drugs,diseases and targets,select the best correlation information after testing and experiment for further processing,and build drug-drug association network,drug-disease association network and diseasedisease association network on this basis.A new multimodal graph is constructed based on multiple association networks.The multi-layer graph convolution neural network architecture is used to learn the characteristics of the heterogeneous network,and the attention mechanism is used to integrate all useful structural information from multiple graph convolutions.Finally,a well-defined scoring function based on integrated embedding is used to give the predicted score of the unobserved association between drugs and diseases,and then make clinical drug decisions.For the first time,the graph convolution model based on attention mechanism was used to fuse the correlation information of drugs,diseases and targets.The system can provide clinical medication suggestions according to patients’ diseases.In order to verify the effect of the system prediction,the obtained drug-disease association was verified by literature,instructions and clinical trials to verify the effectiveness of the model.Results: After experimental comparison,the drug-drug association network is constructed by selecting the drug treatment similarity association information,finetuning the parameters of the association network through the scoring function,and constructing three kinds of association diagrams based on the efficient and accurate extraction of the association information,and using the deep learning multi-layer graph convolution network architecture based on the attention mechanism for learning,Finally,"Deep Graph Clinical Decision Support System(DGCDSS)" was obtained.Including732 drugs,440 diseases and 1915 human targets as the whole information set.Finally,when the minimum train-loss is 0.54814,the prediction results are output,with an average accuracy of 98.59%,an accuracy recall curve(AUPR)score of 0.3117,and an area under the receiver-work characteristic curve(AUC)score of 0.8887.The results obtained were further analyzed with an average verifiable accuracy of 64.2%.Conclusion: The multiple correlation network selected in this study has high accuracy,precision recall curve(AUPR)score and receiver work characteristic curve area(AUC)score,which can fully extract the correlation information of drugs,diseases and targets.The graph convolution neural network based on attention mechanism is used for the first time for the fusion of information related to drugs,diseases and targets,with high accuracy.The predicted results can effectively improve the research and development efficiency of new old drugs and provide doctors with new treatment methods.