Construction And Prediction Of Microbes-Diseases Associations Knowledge Base Based On Text Mining

Microbes are everywhere in nature.These microorganisms,especially bacteria,live on the human body,regulate the balance of the body and affect the metabolism through interaction.In recent years,with the rapid development of experimental conditions and technological level,the research data on microorganisms have been rapidly expanded.The accumulation of these data has led to the emergence of genomic sequence analysis,biological network construction,biomedical text mining and other computing methods,which are used to explore the potential knowledge in the massive biomedical data.This study mainly focuses on the extraction of bacteria-disease association from the perspective of biomedical text mining,with the usage of entity recognition,relationship extraction,knowledge base construction,association prediction and other methods to study on microbes and human health.The following shows the main work of this thesis:(1)To propose a Wipedia-based method of bacteria and disease association mining.Previous studies have built databases of bacteria and disease through literature search and extraction,except Wikipedia,the world's largest public knowledge platform.Wikipedia contains such an enormous associations between bacteria and disease entities that it might be a good complement to literature mining.To address this issue,this article used Kindred,a text mining tool,to extract the interactions between bacteria and disease from the texts in Wikipedia.This work expands the existing database of bacteria and disease by about 16 percent,proving that text mining via Wikipedia is a powerful complement to literature mining.(2)To develop a Django-based bacteria and disease interaction visualization platform.Existing databases related to bacteria and disease only consider the simple correlation information,resulting in the function limitation.In this study,the multi-source attribute information such as bacterial habitat and host is integrated,and the correlation between bacterial attribute and disease can be verified by analyzing the correlation between bacterial attribute and disease.This system platform also realizes the query and visualization of correlation data,which provides a more intuitive and convenient tool for visualization and reliable evidence for future correlation prediction analysis.(3)To propose a correlation prediction method based on knowledge base of bacteria and disease.The association prediction analysis from the perspective of text has been widely used in the general fields.However,the application in the field of biomedicine,especially the microbial association,is still rare.This thesis proposes a method to predict the association between bacteria and disease based on bilinear model,neural network and translation model.The results show that the predictive effect of translation representation learning is better than the other methods,which can predict unknown associations quickly and effectively and realize the function of knowledge base completion.In this thesis,the research and visualization platform for knowledge base of bacteria and diseases can realize the tasks of knowledge mining and relationship prediction of bacteria and diseases,and provide an intuitive and convenient visual analysis tool.