Analysis And Recommendation Of Rational Drug Use Based On Knowledge Graph

Rational drug use plays an important role in drug discovery and clinical treatment.Drug discovery and clinical prescription rely primarily on the experience and expertise of medical personnel,which are often time-consuming and risky.Nowadays,owing to more and more pharmacal resources in the medical field,it is of great significance to effectively explore the potential knowledge of drug which promotes drug development and personalized medicine.At present,artificial intelligence is used for drug reuse prediction and drug recommendations for new diseases,but these methods often ignore external knowledge that is not mentioned in the training data.Therefore,in this thesis,a new drug knowledge graph is designed by integrating the existing pharmacological databases,which could provide feasible solutions for drug reuse prediction and drug recommendation with deep learning methods.The speficic research results are as follows.(1)A drug knowledge graph was constructed with the fusion of multi-source semantic information.Relationships between Disease-Drug,Disease-Target and DrugDrug are extracted from pharmacological databases such as Drugbank and Pub Chem,and the semantic information of various pharmacological databases was fused with medical language system.A total of 17,834 drug entities and 180,543 drug-related entity relationship pairs were collected to intuitively display multidimensional entity relationships,and the Trans E model was used for embedded analysis,thus providing a priori knowledge for the clinical drug recommendation.(2)A drug target binding affinity prediction model combining pre-training and multi-task learning was designed and implemented.The model uses Transformer and Graph Convolutional Networks(GCN)to obtain the feature representation of drugs and targets respectively.Aiming at the over-fitting problem caused by the small amount of data of drug target pairs and the large number of parameters of the deep learning model,a multi-task framework of dual adaptaion mechanisms combined with pre-training was proposed,which is validated experimentally on the standard drugtarget binding affinity dataset DAVIS.The mean square error of the proposed model is0.847,which is 17% higher than the benchmark performance,and the results were analyzed by ablation experiment and the replacement of pretraining method.It indicaties that the proposed method alleviates the over-fitting problem and prospectively discover new drug-target relationships.Finally,the model was used to predict the interactions between the two kinds of the Corona Virus Disease(COVID-19)targets in the authoritative drugs of Food and Drug Administration(FDA)of the United States in order to find the effective drugs that may be targeted at COVID-19.In addition,new potential drug-disease relationship was found through drug reuse to complete the drug knowledge graph and enrich the construction of medical knowledge system.(3)A combined drug recommendation model combining knowledge graph and deep learning was designed and implemented.For clinical data,the Transformer model was used to extract features and predict candidate drug combinations.To solve the problem of over-fitting model due to the large number of deep network parameters,the model was pre-trained by using a single diagnosis and treatment record.In view of the risk of drug combination,drug-drug interaction in knowledge graph was used to constrain the candidate drug combination.The model was verified on public mimic data.The experimental results showed that the Jaccard similarity coefficient of the model increased by 16% compared with the baseline,and the experimental results were analyzed by Convergence curves and ablation experiments.It is providing technical support for clinical rational drug use.In summary,taking COVID-19 disease effective drug analysis recommendation as an example to study rational drug use analysis and recommendation method andaiming at the problems of DTA prediction and irrational drug combination of drugs which were neglected in training field,the drug knowledge graph,drug target binding affinity prediction model and drug combination recommendation model were designed and constructed.On this basis,the prototype system of clinical rational drug use recommendation based on knowledge graph was designed and implemented,which provided beneficial exploration for clinical medicine to provide reasonable drug reuse solutions,and provided systematic support for intelligent drug development and personalized medical service.