Research On Computational Drug Repositioning Based On Multi-information Fusion

The traditional drug research and development of pharmaceutical enterprises face some problems,such as long period,huge investment and high risk.Drug repositioning aims to find new indications of existing drugs,that is to say,“new uses of old drugs”,which can effectively shorten the drug research and development cycle,reduce the costs and risks.It has become an important drug development strategy for pharmaceutical companies.Generally,some early cases of drug repositioning came from accidental discovery or experimental screening.With the accumulation of large amounts of drugs and diseases biomedical data and the development of computing technology,computational drug repositioning can provide meaningful guidance for drug research and development,and has become a hotspot in the research field of computational biology.Although many computational drug repositioning methods have been proposed,however,how to effectively integrate various biological data and construct computational model with high prediction accuracy has been a major aspect of researchers.The main research work of this thesis is as follows:(1)Considering the effect of the topology information of the drug-disease association network on prediction performance,a drug repositioning method based on integrated drug disease topology information and Bi-Random walk(Bi RW)algorithm is proposed.First,the drug-drug and disease-disease topological similarity is measured based on a deep learning method Deep Walk algorithm.Then,some drug or disease similarity are integrated,which contains the drug topological similarity,disease topological similarity,drug chemical structure similarity,disease phenotype similarity,drug Gaussian kernel similarity and disease Gaussian kernel similarity.After that,the integrated similarity is further adjusted to obtain an improved drug similarity and disease similarity.On this basis,a drug-disease heterogeneous network is constructed and Bi RW algorithm is adopted to obtain potential drug-disease associations.The experimental results and case studies on the standard dataset demonstrate that the algorithm shows superior prediction performance.(2)In view of the sparsity of known drug-disease association information,considering the disease or drug entity neighbor information,a new similarity fusion strategy and two-step propagation method is proposed to predict potential drug-disease association.First,WKNKN method is used to improve the drug-disease association matrix and obtain the drug cosine similarity and disease cosine similarity.Then,the entropy values of different similarities are calculated.On this basis,multiple drug similarity and the disease similarity are fused and construct new drug-disease association matrix.Finally,a two-step propagation method is used to predict the potential drug-disease association.The experimental results and case studies on the standard dataset demonstrate that the algorithm shows superior prediction performance.(3)Considering the advantages of different drug repositioning methods in discovering potential drug-disease association information,a drug repositioning method based on ensemble learning is proposed.First,the sparse drug-disease association matrix is pretreated and obtain the drug linear neighborhood similarity and disease linear neighborhood similarity.Then,the different prediction models are analyzed,and the prediction results of these prediction models are further integrated.The experimental results and case studies on the standard dataset demonstrate that the algorithm shows superior prediction performance.