Microbe-Host Attributes Association Prediction Based On Heterogeneous Network

With the development of bioinformatics,a great deal of microbial omics data has been produced.Researchers have found that the composition and structure of microbial communities are closely related to human diseases.To further explain the pathogenesis of complex diseases such as diabetes,obesity,depression and so on.The study of microbe-disease association has become a hot topic in bioinformatics,it's challenging to study the microbe-disease association owing to the complex structure of the microbial community.Microorganisms are mainly composed of bacteria,fungi and viruses.The viruses infect the human body by the proteins of the virus interacting with the proteins of the human body,which affects the metabolic function of the healthy human body and leads to human diseases.We can understand the disease mechanism from a molecular perspective by studying virus-human protein-protein interaction.The main research work and innovation are as follows:First,Bi-Random Walk based on Multiple Path to predict microbe-disease association.The microbial network was inferred by abundant data in HMP,and disease similarity was built based on disease symptoms,then microbe-disease heterogeneous network was constructed combining with known microbe-disease associations.We proposed Bi-Random Walk based on Multiple Path algorithm by analyzing the principle of random walk.We rank the disease-related microbe according to the association strength with disease to infer the potential microbe-disease association.Finally,the performance of the algorithm is proved to be better than other algorithms in this field through parameter adjustment and cross-validation.We also prioritized potential microbes related with Obesity,Asthma and Liver Cirrhosis three common diseases and demonstrated its reasonableness through looking up medical literatures.Second,predicting virus-human protein-protein interaction based on Network Consistency Similarity with Harmonic Mean.Considering the sparsity and incompleteness of existing virus-human protein interaction data and the principle of Network Consistency projection algorithm,and we proposed Network Consistency Similarity with Harmonic Mean algorithm.The virus protein network is as important as the human protein network in this algorithm,and we use harmonic mean for two scores which get by virus protein network and human protein network as the final virus-human protein-protein score.The performance of NCS_HM proposed in thesis is better than NCP,OVPN,OHPN,LP and PMF algorithms,which proved that the performance of a heterogeneous network model is better than a single network.We use the row or column of virus-human protein-protein interaction cross-validation to prove the performance of the algorithm in predicting isolated proteins.Finally,we use DAVID function annotate tools on experimental and predicted protein sets and verify the reliability of predicted virus-human protein-protein interaction from the biological view.