Computational Methods For Extracting Higher-order Microorganism Interaction Based On Probability Logic

With the rapid development of Human Microbiome Project(HMP),the correlation between microorganisms has become a significant research hotspot in the field of bioinformatics.Utilizing second-generation sequencing technology,researchers are able to obtain sufficient microbial community and functional gene data.Although many methods have been proposed for analyzing the microbial interactions and a lot of experiments have been conducted,most of works only focus on the relationships between two species.In this thesis,we use a probabilistic logic-based approach to infer high-order interactions of microbes,break through the limitation of the paired relationship,study the complex relationship among the three microorganisms,and systematically apply it to the data of the human microbiome project.Based on the obtained high-order interactions of microorganisms,a high-order interaction network of microorganisms can be constructed based on the three uniform hypergraphs.We study the method of keystone species identification and sample classification according to the characteristics of hypergraphs,and achieve good results.The main research work and contributions are as follows:(1)Higher-order interaction extraction of microorganisms based on probability logic.In this thesis,we use the method based on probabilistic logic to mine the high-order interactions of microorganisms.4367 higher-order logical relationships of microorganisms are found on the data provided by the Human Microbiome Project,which proves the universal existence of higher-order logic.We also compare the distribution of higher-order logic types in different parts of the body.Compared with the classical association rules algorithm,the experimental results show that the probabilistic logic method is more effective than the existing association rules method.Not only can high-order interactions between high-abundance microorganisms be excavated,but also high-order interactions between a large number of low-abundance microorganisms.Moreover,it can overcome the shortcomings of a large number of redundant rules and excessive memory overhead in association rules.By analyzing the metabolic network of microorganisms,we find that the resource interactions among microorganisms contribute to the higher order interactions,which explains the biological significance of the calculated results.(2)Keystone species recognition of microorganisms and sample classifiction based on the analysis of three uniform hypergraphs.Most of the current microbial interaction networks are simple networks with paired associations,while the hypergraphs based on higher-order interactions can better visualize and describe the complex interactions of microorganisms.Based on the obtained high-order relationship,the interaction network-three uniform hypergraph was built.And a degree and loss degree of hypergraph nodes hybrid method was proposed to identify key species in microbial high-order networks.In order to verify the validity and availability of the method,253 species of microorganisms of the genus level were labeled as key species or non-critical species.The marked data set is divided into training set and test set according to 1:1,and the classification is performed by supervised LDA discriminant model.Then the characteristics of the hypergraph,including betweenness centrality,closeness centrality and eigenvector centrality were calculated.The experimental results show that the characteristics based on hypergraph can effectively classify the sample,and the accuracy rate reaches 90.61%.