Research On Drug Redirection Method Based On Network Representation And Learning To Rank

Drug redirection analysis is a technology used to find out the indications of existing drugs or drugs under research that are beyond the original approval,and to expand their scope of application and use.Compared with traditional drug research and development,this technology can greatly save the labor and material costs of drug research and development.However,there are still challenges in drug redirection: 1)Due to the complexity and diversity of data,there is a lack of uniform standards and norms for constructing drug redirection data sets;2)Because the same drug has different effects in treating different indications,most drugs It is not particularly appropriate for the redirection method to regard it as a qualitative forecasting problem.By exploring the influence of different drug characteristics on drug redirection,this study constructed a higher performance drug redirection method.Through the combination of literature mining and manual review,this study constructed a data set of drug efficacy comparison,and established a quantitative prediction method of drug redirection combining network embedding and ranking learning.This article mainly includes the following work:(1)Evaluation of drug classification system based on network pharmacological mechanism.Drug redirection discovers new indications through the calculation and analysis of drug characteristics,reflecting the similarities and differences between drugs.The most direct feature of a drug is its indication.The current international standard for classifying drugs through indications is the Anatomical Therapeutic Chemistry Code(ATC).Therefore,calculated the drug similarity under different feature data,compared and analyzed the difference in drug similarity between actual and random conditions,and obtained the relationship between the different characteristics of the drug and the ATC and the rationality and deficiencies of the ATC itself,which provides for the construction of the drug redirection data set.It is an important basis for feature selection and provides support for proposing a new drug classification system.(2)Research on drug redirection method based on heterogeneous network.Aiming at the problem that the network information of homogeneous network or bipartite graph structure is relatively single,this study proposed a method of drug redirection based on heterogeneous network.First,by comparing heterogeneous networks that integrate different information,using different network embedding representation methods and knowledge complement methods,comprehensively evaluated the drug redirection performance of existing methods under different heterogeneous networks.Based on this,a drug redirection method combining embedding representation and convolutional neural network(Con EE＿DR)is proposed.The experimental results show that the performance of Deep Walk is the best among the methods based on network embedding.Among the methods based on knowledge completion,Conv E has the best performance.Compared with these two methods,our proposed Con EE＿DR method has improved prediction performance by 30.8% and 13%,respectively.The top 10 hit rate and MRR reached respectively.0.605 and 0.350.(3)Research on drug redirection method based on ranking learning.In order to solve the problem of drug redirection from a quantitative perspective,this study used a combination of literature mining and manual review to construct a drug efficacy data set that includes the relationship between drugs and indications,and proposed drug redirection computing framework DRRNet that combine network embedding and ranking learning..Experimental results show that our proposed DRRNet method achieves an NDCG@10 of 0.9278.At the same time,compared to the best method based on network embedding,the hit rate Hit@10 is increased by 27.14%;compared to the best method based on knowledge completion,the hit rate is Hit@10 increased by27.45%.