A Research On The Prediction Model Of Microbe-disease Associations Based On Biological Network

With the rapid development of high-throughput sequencing technology and modern bioinformatics,the study of microbiology has received increasing attention from the scientific and medical communities.Studies over the years have demonstrated that microorganisms are closely related to human diseases.A deeper insight into the correlation between microbes and diseases will not only reveal more about the etiopathogenesis of human diseases,but also provide new perspectives on the prevention and treatment of diseases,thus promoting global human health.However,traditional biologically validated experiments are both time-consuming and laborious.Therefore,the use of computational methods to predict potential microbe-disease associations has become a new research topic in bioinformatics.In this thesis,we focus on designing microbe-disease association prediction models using known microbe-disease association data using network algorithm and machine learning approach respectively,with the following main tasks:(1)Collection and processing of biological data required for the model.A large amount of validated microbe-disease association data was collected manually,and a new microbedisease association database MDADP was constructed,containing 1172 microbe-disease association relationships between 51 diseases and 473 microbes.On this basis,the database was filtered and processed to construct a new microbe-disease association dataset.The MDADP database not only provides a new dataset for the field of microbial-disease association prediction,but also promises to uncover more novel and critical biomedical information in future studies.(2)The microbe-disease association prediction model BWNMHMDA based on bidirectional weighted biological networks is proposed.Firstly,three association networks are constructed based on microbe-disease association data and Gaussian Interaction Profile Kernel similarity,and a microbe-disease heterogeneous biological network are obtained by integrating these three association networks.Secondly,after assigning weights to each node in the biological network,normatively adjusting the weights of each edge in the network,and introducing a unique two-way recommendation algorithm,the microbe-disease heterogeneous biological network with two-way weights is obtained.Finally,based on this bi-directional weighted heterogeneous biological network,a new matrix calculation method is proposed to infer the potential microbial-disease associations.(3)A microbe-disease association prediction model BPNNHMDA based on backpropagation biological network is proposed.Firstly,a unique back-propagation biological network is constructed with the input data as a matrix of microbe-disease associations and the output data as a probability matrix of potential microbe-disease associations.In addition,an improved activation function based on hyperbolic tangent function is designed for the activation of different network layers in BPNNHMDA,and the initial edge weights of the network are optimized by Gaussian Interaction Profile Kernel similarity,which effectively improves the training speed of BPNNHMDA.In this thesis,the implementation of the model's performance is assessed by leave-oneout cross-validation,K-fold cross-validation and combined with case studies.The experimental results showed that the proposed two prediction models have superior and reliable prediction performance compared with other leading-edge methods.This suggests that the two prediction models can provide valuable potential microbe-disease associations for future biological experiments.The thesis concludes with a summary of the research work and an outlook for future research.