Study On Associations Between Complex Diseases And MicroRNAs,microbes

MicroRNA(miRNA)is a type of endogenous,noncoding single-stranded RNA molecules with about 23 nucleotides.Since the first discovery in the 1990's,miRNAs have ushered in a new field of molecular biology and is also an important frontier in life science in the post-genome era.With the rapid development of next generation sequencing and biological experiment technology,cumulative reaserches have continuously confirmed that the dysfunction of miRNAs is closely related to the emergency and development of human complex diseases such as malignant tumor,cardiovascular disease,nervous system disease and so on.In addition,as the Human Microbiome Project(HMP)enters the second phase(HMP2),as an important part of human ecosystem,the imbalance of microorganisms has been proven to cause a variety of diseases.Therefore,the research about the association of complex diseases with miRNAs and microbes is of great significance for understanding the pathogenesis of diseases and also the prevention,diagnosis,treatment and prognosis.The main content of this article is that based on multi-source biological data information,we design some efficient and reliable mathematical prediction models and algorithms to explore the potential associations between complex diseases and miRNAs or microbes.We are aimed to provide promising verification canditates for biological experiments as a priori to improve experiment efficiency and shorten the discovery cycle.In addition,we propose an information spreading model based on the structural balance theory in signed network,which could simulate positive and negative relations,providing a theoretical basis for furher research on the association prediction of signed biological networks.There are five chapters in this article.The introduction is in the first chapter,which briefly introduces the relationship and research progress of miRNA,microorganisms and diseases.In the second and third chapters we respectively introduce five computational models,which are proposed for the association prediction of complex diseases and miRNAs,microbes.We introduce the theoretical basis and framework of the models,as well as the evaluation and analysis of prediction performances.In chapter 4,we introduce the information spreading model under the influence of the structural balance theory in signed networks,including the theoretical basis and analysis of simulation and empirical results.In chapter 5 we summarize the works done,and make plans and prospects for future research.Firstly,in the study of the association between complex diseases and miRNA,we designed four mathematical models to predict reliable potential associations.1.Considering that in previous research,many measures about similarity between miRNAs and diseases have been provided taking use of various mathematical methods based on different biological processes.It is difficult to have a unified standard to judge which is good or bad for such a large number of indicators.Therefore,we design a multi-kernel learning optimization iterative method to optimize the combination of existing multi-source similarity information.In the situations where many indicators are involved,our model could give the best way for the similarity combination.Based on the Kronecker regularized least squares classifier,we use the best combination of similarity results to predict the association between unknown diseases and miRNAs,and obtained the most robust and accurate prediction results at the time.2.We propose a calculation model of miRNA and disease association prediction based on random walk and binary regression.In this model,we give a methodology to extract feature vectors of arbitrary dimensions by using random walks on graph.The characteristic of this model is that the setting approach for feature dimension can be combined with domain prior and data information,also the complexity of the binary regression model is very low.Thus,this model could be adapted to prediction task in many other fields.3.The traditional random walk process usually treats neighbor nodes equally and transfers them with a uniform probability.When we integrate the relation between miRNA and complex diseases into a heterogeneous network,there will be differences between the neighbors of the node.The traditional random walk method will no longer be applied.Therefore,we propose an algorithm based on the maximum entropy random walk to carry out potential association predictions.In the maximum entropy random walk,the transfer process of the random walk will reach the greatest randomness.In the model,we specifically give the entropy rate calculation method of the random walk process,and analyze that the maximum entropy random walk method is consistent with the eigenvector centrality centrality of the network topology.Thus,it could distinguish the importance of miRNA nodes and disease nodes.4.Considering that the association prediction problem between miRNAs and diseases is essentially a recommendation problem,we propose a hybrid graph recommendation prediction algorithm.In the model,the resource allocation process in bipartite graph is used to construct a graph projection about miRNA and disease,at the same time,feature integration is performed based on the common neighbors of network nodes,after which prediction tasks are performed in the miRAN space and the disease space respectively.Thereafter,in the study about the association between complex diseases and microbes,we design a model of random walk on hypergraph to predict potential microbe-disease associations.In this model,we introduce the hypergraph approach to model the relationship between diseases and microbes.In the hypergraph model,any edge can carry all the known information related to the object,meaning that a hyperedge comprises many nodes.Compared with the ordinary graph model,it not only avoids the limitation of information loss,but also provide interpretability at the biology level.In addition,we propose a hypergraph random walk algorithm,in which we integrate multi-source similarity data to estimate the transition probability.From the perspect of biology,we provide an adaptive strategy for distinguishing the importance of miRNAs to ensure the accuracy and stability of prediction.Finally,since the existing models hardly consider the positive and negative effects and relationships between miRNA and complex disease networks,such as the effects of miRNA up-regulation and down-regulation on disease development,and few models simulate the positive and negative effects between biological factors,such as the positive and negative regulatory relationship between genes,the enhancement and antagonism in drug combination,and so on.We,for the first time,propose an information spreading model based on the structural balance theory in signed networks,which is aimed to provide some theoretical foundations for solving this kind of problems.There are two different kinds of edges in signed networks,namely positive edge and negative edgeg,which can be used to describe the positive or negative regulation in biology networks.In particular,based on the shortest path and structural balance theory,we develop an algorithm for identifying potential relationship in signed networks.Simulations on random signed networks and empirical experiments on real datasets show that our proposed information spreading process also can be approximated by local 2-order neighborhood.In addition,we found that the ratio of the number of potentially positive nodes in the network is consistent with the network content.Our model can be applied to a wide range of real scenarios,such as gene regulation prediction,drug combination prediction and so on.Our model provides a theoretical foundation for further research on prediction problems about signed biological networks.