Research On The Prediction Model Of Disease-related Potential Biomarkers

Ideal biomarkers have very important value and function for the early diagnosis,identification and monitoring of diseases.In the past few years,more and more evidences have proved that microorganisms and lncRNA play a crucial role in the basic life activities of the human body.Therefore,microorganisms and lncRNA are considered as potential biomarkers and have received extensive attention in human disease research.However,traditional biological identification experiments are costly and time-consuming.Therefore,in order to advance the identification process of biomarkers,researchers began to use more efficient computational models to infer the relationship between microorganisms and diseases,as well as lncRNAs and diseases.This thesis introduces the prediction problem of disease-related biomarkers,and develops computational models to conduct specific research on the relationship between diseases and two types of potential biomarkers,microorganisms and diseases.The main contents are as follows:(1)For the prediction of microbe-disease association relationship,a prediction model based on two-way label propagation,NBLPIHMDA,is proposed.In NBLPIHMDA,the Gaussian kernel function is first introduced based on the known correlation distribution between diseases and microorganisms,and then the similarity between microorganisms and the similarity between diseases are calculated.Secondly,according to the similarity of these two types of nodes,the microbe weighted network and the disease weighted network are constructed.Subsequently,by integrating these two weighted networks with the known microbe-disease association network,a microbe-disease heterogeneous network was constructed.Finally,based on the heterogeneous network,a two-way label propagation algorithm is implemented to calculate the correlation scores of all unknown microbe-disease pairs.In addition,various experimental results show that NBLPIHMDA has reliable performance.(2)For the prediction of lncRNA-disease association relationship,a local radial basis biological network model ICLRBBN based on internal confidence is proposed.In ICLRBBN,first,based on the obtained lncRNA-disease association data and the semantic DAG map of the disease,an internal confidence collaborative filtering recommendation algorithm is proposed to mine the indirect features in sparse related data,so as to better predict the lncRNAs associated with new diseases.Then,a three-layer radial basis function network with local characteristics composed of diseases and lncRNA is constructed.Finally,on the network,the correlation scores of all unknown lncRNA-disease pairs are calculated by combining the characteristics of different lncRNAs and diseases.In addition,various experimental results demonstrate that the prediction performance of ICLRBBN is significantly superiorer than previous models,which means that ICLRBBN can provide valuable reference for future biomarker research.Finally,this thesis analyzes and summarizes the above two prediction models,and points out the direction of improvement in the next step.