Research On Prediction Method Of Drug Disease Relation From Network Perspective

As the cost of new drug development increases and the number of approved drugs decreases significantly,the market's need for identification of drug targets and their corresponding innovative efficacy prediction methods is increasing.Moreover,drug development for complex diseases faces many challenges,including insufficient efficacy,serious drug resistance,adverse drug reactions,incomplete evaluation,and personalized drug use.At the same time,drug development has shifted from the traditional single-target model to the multi-target model based on complex living systems.In this context,we propose a new disease-drug prediction method from the perspective of gene-gene network,disease-gene network and drug-target network,which promotes the research of drug relocation to some extent.Specifically,the main research contents of this paper are as follows:First of all,starting from the innovation of research,we propose a new network research method.From the perspective of gene network,nearly 20,000 kinds of genetic data are regarded as nodes in the network,combined with BFS and DFS for random walk to select the next node,and more fully discover the known information in the gene network for each node.Generate the corresponding node sequence,and then use the Word2 Vec model to implement vector representation for each node,and then use Euclidean distance to define the similarity between genes.The so-called complex network combined with machine learning methods to study gene-gene proximity,and on this basis to quantify the relationship between drugs and disease.In the process of drug-disease similarity calculation,based on the measure of similarity between genes,we propose a new method to efficiently predict the drug efficacy,namely vector distance calculation.By constructing the reference distance distribution to reduce the error of drug prediction,and comparing the method of calculating the distance with gene degree in the classical method,it is verified that the method of using vector can better screen the drug.At the same time,we also proved that the shortest distance calculation method is better than the average distance calculation method,that is,the actual effective drug and disease Z-Score is smaller.The dcv vector nearest distance calculation method was used to analyze the drug prediction results of two common diseases of arrhythmia and diabetes.The final results were in good agreement with the data in the medical database.The last special case analyzes epilepsy,a common disease that can treat more drugs,and proves that there may be a relationship between zolpidem and epilepsy,which is worthy of further verification by offline medical experiments.The whole experiment results show that the gene is used as a new perspective of research,using machine learning model to analyze the characteristics of gene network,and using vector nearest distance calculation method to predict the relationship between drugs and diseases,the results corresponding to a larger AUC value,that is,higher accuracy.Finally,combined with the practical application,the drug recommendation information system was designed to help medical personnel complete the corresponding information inquiry and realize the development of new drugs.