Study On Prediction Model Of Associations Between Disease And Biomarker

Since the ideal biomarkers is effective for reflecting the structural change or functional change of organ,tissue,system and cell,they play a key role in the prevention,diagnose,monitoring,prognosis and treatment of disease.In recent years,the increasing research suggests that microRNAs and microorganism have a great effect on the involvement of basic biological processes,including pathogenic effect.Therefore,seeking potential ideal biomarkers from perspectives of micro RNAs and microorganism becomes a hotspot issue.However only depending on biological experiments to explore biomarkers is expensive and time-consuming.As diverse types of biological datasets open access,researchers have utilized computational models to predict potential mi RNA-and microbe-disease association relationships from various perspectives.It can prioritize the most likely candidates which are preferentially experimentally validated.As a consequence,it not only boosts the identification of high quality biomarkers but also enhances the understanding of mechanisms of complex human diseases.The main contributions of this work can be summarized as follows:(1)The biological datasets were collected,cleaned and preprocessed.Based on the processed datasets,disease semantic similarity,mi RNA functional similarity and Gaussian interaction profile kernel similarity were calculated.(2)This dissertation proposed a Path-Based computational model for Mi RNA-Disease Association prediction named PBMDA.It constructs a bicolor heterogeneous network and then adopts a special depth-first search algorithm to traverse all paths in the graph for scoring all potential associations.Those ones with high scores could be considered as the most likely candidates.(3)This dissertation proposed Laplacian Regularized Least Squares for Human MicrobeDisease Association prediction named LRLSHMDA.It first implements Laplacian normalization operation and minimizes the cost functions for optimal classifiers in microbe and disease space.Finally,those two classifiers are transformed into an integrated classifier for calculating the probability values of potential association relationships.The prediction results were validated by leave-one-out and 5-fold cross validation and case studies.The validation result fully demonstrated the more reliable prediction performance of these two computational models than other state-of-the-art models.It is anticipated that this work could provide an insight into the future related research.