Drug-Disease Interactions Prediction Based On Similarity Method

New drug research and development costs are large and development time is long.So,researchers hope to develop a new method of drug discovery and shorten the development cycle while providing a good income.Drug repositioning methods can speed up the drug discovery process,so it is getting more and more attention from researchers.In addition,with the increasing data on drugs and diseases,it is impractical to use experimental methods to verify all drug-disease associations.Researchers have developed a number of computational methods to provide assistance and reference for experimental validation.Studies have shown that the integration of multiple information helps predict drug-disease interactions,but how to efficiently integrate information from different sources remains a challenging issue.Previous prediction methods still have drawbacks in predicting performance.This paper focuses on how to predict drug-disease interactions by effectively integrating data from multiple sources,so as to realize the prediction of drug repositioning and indication of novel drugs.The main work is summarized as follows:First,a drug-disease interaction prediction method based on similarity network analysis is proposed.Firstly,multiple drug similarity matrices and disease similarity matrices were constructed using the Jaccard similarity coefficients based on collected data from multiple sources.Secondly,different drug similarity networks and disease similarity networks are established based on the obtained similarity matrices and a method called similarity network fusion methods are used to integrate the drug similarity networks and disease similarity networks of various sources.Finally,the integrated drug similarity network and disease similarity network are mapped to a large cyberspace through known drug-disease interactions and random walks are used to predict unknown drug-disease interactions.Experimental results show that the prediction performance of the model is improved.Second,a drug-disease interaction prediction method based on integrated learning is proposed.Firstly,multiple sets of drug-disease pairs sample features were constructed using the feature construction methods presented herein based on data from multiple sources collected.Secondly,multiple classifiers were constructed using sample characteristics for multiple sets of drug-disease pairs.Finally,the idea of integrated learning is used to integrate multiple classifiers to build an integrated model to predict unknown drug-disease interactions.Cross-validation results showed that the model performed well in drug relocation and prediction of new drug indications.The case study results show that the model has certain practicability.