Study On The Construction Of Knowledge Map Of Anti-tumor Drugs And Personalized Drug Recommendation

Objective:The increase of tumor incidence rate has always threatened human life and health,and the complexity and heterogeneity of tumors make tumor treatment face enormous difficulties,and the mortality of tumor patients remains high.For the treatment of tumor,traditional surgery,radiotherapy and chemotherapy have great side effects on human body and poor prognosis.In recent years,targeted drug therapy,based on the big data of clinical treatment information and gene mutation information,adopts a variety of high-throughput detection methods to accurately attack the cancer sites identified by tumor cells at the cellular and molecular levels without damaging the normal tissue cells around the tumor,and truly realizes personalized drug use,which shows that accurate personalized drug use plays a central role in the anti-tumor process.This paper proposes a recommendation prediction model based on the knowledge map of anti-tumor drugs and the structural data of multi-modal graph constructed by using graph convolution neural network deep learning,which can be used for personalized recommendation of anti-tumor drugs.Method:This paper studies and develops a drug recommendation prediction model based on the knowledge map of anti-tumor drugs and the structural data of multi-modal graph constructed by using the graph convolution neural network deep learning to realize the personalized recommendation research of anti-tumor drugs based on targets.Use crawler technology to crawl Drugbank,DDinter and PPI databases to extract drug and target information.Use Neo4 j to build a knowledge map database,and build an anti-tumor drugtarget,drug-drug,target-target association network on this basis.These networks with different modes jointly construct the data of "multimodal graph" structure.Next,the graph convolution neural network algorithm is used to learn the characteristics of "multimodal graph",obtain the feature association network of drugs and targets,and conduct personalized anti-tumor drug recommendation.The traditional graph convolution neural network can only deal with a single graph structure feature.The Decagon model adopted in this paper can break through the traditional limitations and simultaneously deal with and learn multiple different types of graph structures.The graph convolution neural network algorithm model in this study is compared with the traditional link prediction results to verify the effectiveness of the model.Results:After knowledge fusion,715 drugs,2834 targets,19442 drug-drug associations,12718 drug-target associations,and 96712 target-target associations were finally obtained.The drug-target knowledge map was constructed through Neo4 j for visual presentation.The constructed knowledge map used MATCH in Cypher language for convenient query and search of drug-drug,drug-target,and target-target relationships.The construction of the knowledge map provides the basis for the establishment of the multimodal graph structure data.After learning the Decagon model,the new multimodal graph structure is reconstructed for drug prediction.The "connectivity" of the target is used as the index of personalized drug recommendation,and the possible therapeutic drugs of each target are comprehensively evaluated with the evaluation index of accuracy and recall,The top 10 drugs with the best performance of AUPRC and the 10 drugs with the worst performance of AUPRC are listed according to the predicted probability score.The accuracy rate of the top five target points and the top three drugs with the recall rate are compared as the final drug recommendation results.In this study,we use F1-score to evaluate the performance of model training,and compare the Decagon model with the traditional link prediction model.The result shows that compared with other models,the Decagon model has a higher advantage in the current drug recommendation based on in-depth learning,and the F1-score score reaches 93.72%.The drug recommendation results were used in clinical practice,and the overall survival rate was significantly improved,which can provide clinical treatment suggestions for doctors.Conclusion:This study uses Neo4 j map database to store drug-target knowledge map,uses Neo4 j to visualize knowledge map,and provides data preparation for drug recommendation.The accuracy and recall rate of drug recommendation system based on multimodal graph convolution neural network training is much higher than that of traditional link prediction model.In the future,the depth map structure clinical drug recommendation system can combine the big data of the Internet with the hospital,and can quickly give drug recommendations after the doctor detects a certain gene abnormality of the tumor patient,providing more drug choices for doctors.