【作者】 钟颖莉；

【导师】 李建中；

【作者基本信息】 黑龙江大学 ， 微电子学与固体电子学， 2017， 博士

【摘要】 现代医学研究表明,基因是绝大多数人类疾病的根源。随着人类基因组计划的完成,研究人员已经从全基因组测序过渡转化到基因组功能及其调控机制的深入研究,即从基因的调控功能层面发现人类疾病的奥秘。由于整个人类基因组中能编码蛋白质的DNA序列只占整个基因组的3-5%,而绝大多数的DNA序列仅仅转录成RNA,并不参与蛋白质的直接编码。因此,非编码RNA调控基因表达的作用逐渐显现,成为生命科学新的研究热点。其中,非编码RNA中的microRNA受到研究人员的广泛关注。microRNA是一种长度约为20-24个核苷酸的非编码小RNA,不参与编码蛋白质,但是在决定蛋白质编码DNA表达上发挥重要的调控作用。研究表明,microRNA参与生命进程各个重要环节,一旦它们的表达异常就可能导致多种疾病的发生、发展。因此,本文致力于构建microRNA功能网络,基于功能网络进行microRNA与人类疾病关联预测问题的研究,发现更多的与人类疾病关联的microRNA候选,为生物学家进行生物实验提供必要的保障。在揭示人类疾病的发生机理及其诊断、治疗方法和推动基因药物的研发等方面都将产生巨大的促进作用。本文研究的主要内容包括以下三个方面:1、研究了准确计算microRNA间功能相似性的计算方法。准确计算microRNA间的功能相似性是构建microRNA功能网络的前提。研究表明功能越相似的microRNA,更可能参与相似疾病的进程,与相似的疾病关联,反之亦然。因此,本文提出通过准确度量两个microRNA关联的两组疾病的语义相似性,度量两个microRNA间的功能相似性的计算方法MFSS。首先,根据国家医药图书馆Mesh中的疾病术语信息建立疾病术语的有向无环图DAG。其次,改进原有疾病术语信息含量的计算方法,根据每个疾病术语上位结点及下位结点的个数准确计算疾病术语的信息含量作为其语义贡献。然后,根据疾病术语的信息含量计算两个疾病的语义相似性,继而计算两个microRNA关联的两组疾病的语义相似性。最后,将两组疾病的语义相似性转换为两个microRNA间的功能相似性,并且构建microRNA功能网络。此外,通过microRNA家族内,microRNA分簇内以及非家族非分簇内的microRNA验证了该方法的有效性,并且针对15种人类疾病,与现有microRNA功能相似性计算方法进行对比,结果表明本文提出的方法计算性能较好。2、基于microRNA功能网络,提出了基于随机游走的microRNA与疾病关联的预测方法。在准确计算microRNA功能相似性的基础上,构建microRNA功能网络,将microRNA与疾病关联的预测问题转换为随机游走问题,提出了基于随机游走的microRNA与疾病关联的预测方法MIDP。该方法结合microRNA功能网络中结点的已知信息,利用已知信息将结点分成两种:有标识结点和无标识结点。其中,有标识结点是已经验证与特定疾病关联的结点,无标识结点是目前还没有发现与特定疾病关联的结点。根据两个microRNA间的功能相似性值,结合两种结点的不同转移权重,建立两种结点的转移概率矩阵。游走者等概率的从与特定疾病关联的任意一个有标识结点出发,根据转移概率矩阵在功能网络中随机游走。多次迭代后,随机游走过程收敛,将矩阵中的概率值定义为特定疾病与microRNA结点的关联分值。最后将关联分值较大的无标识结点作为与特定疾病关联的microRNA候选。使用15种人类疾病与microRNA关联的数据,将MIDP方法与现有方法进行了对比,实验结果证明该方法能够有效预测与特定疾病关联的microRNA候选,并且预测性能优越。3、基于microRNA-疾病双层网络,提出基于非负矩阵分解的microRNA与疾病关联的预测方法。目前,microRNA与疾病关联的预测方法几乎都是针对已知有microRNA关联的疾病进行microRNA候选的预测,对于没有任何microRNA关联的疾病无法预测。本文将microRNA功能网络与疾病网络相关联,构建microRNA和疾病的双层网络,提出带有稀疏约束的基于非负矩阵分解的microRNA与疾病关联的预测方法DMPred。该方法融合双层网络中的microRNA功能相似性信息、疾病相似性信息以及microRNA与疾病的关联信息,利用多种信息之间的关联,同时考虑各种疾病的microRNA候选与所有疾病预测的microRNA候选的关联,进行全面预测。此外,为了提高该方法的预测性能,本文融入了microRNA-疾病关联的稀疏特性。针对15种常见人类疾病,通过交叉验证和模拟实验证实DMPred方法针对预测有无microRNA关联的疾病的microRNA候选均具有优越的性能。更多还原

【Abstract】 Modern medical research showed that genes are the root cause of most human diseases.With the completion of the Human Genome Project,Researchers have transitioned from genome-wide sequencing to in-depth studies of genomic function and its regulatory mechanisms.That is,reaserachers found the mystery of human disease from the regulatory function of the genetic level.Because only 3-5% of the entire human genome can encode the protein DNA sequence.The vast majority of DNA sequences are only transcribed into RNA,but they do not participate in the direct coding of proteins.Therefore,the role of non-coding RNA-regulated gene expression appears gradually and becomes a new research hotspot in life science.Among them,microRNA in noncoding RNAs is widely studied by researchers.Micro RAN is a set of short non-coding RNA with a length of about 20-24 nucleotides.It is not involved in coding proteins,but play a critical regulatory role in determining protein-encoded DNA expression.Accumultaing studies have demonstrated that microRNA participates in all important aspects of biological processes.Furthermore,the abnormal expression of microRNA is one of important causes which result in the occourence and development of various diseases.In this thesis,we are committed to constructing microRNA functional network based on bioinformatics methods,and mining hidden associated information in network by network model,and researching on the disease-related microRNA,and then provide the necessary guarantee of biological experiments for biologists.It will have a tremendous contribution to the diagnosis and treatment of human diseases and the promotion of the development of gene drugs.The creative work mainly consists of the following three parts.(1)The method of accurately calculating the functional similarity of microRNA pair is proposed for constructing microRNA functional network.Accurate calculation of functional similarity of microRNA pair is a prerequisite for constructing microRNA functional network.It is known that microRNA with similar is normally associated with similar diseases and vice versa.Therefore,a method for calculating the functional similarity of microRNA pair,MFSS,is proposed by accurately measuring the semantic similarity of two groups of diseases associated with microRNA pair.First,we construct the directed acyclic graph(DAG)on disease terms according to the information content of disease terms in the national medical library Mesh.Second,we accurately calculate the information content of disease terms as its semantic contribution on the basis of the number of descendants of each disease term to improve the original calculation method on information content of disease terms.Third,the semantic similarity of the two diseases is calculated according to the information content of the disease term,and then the semantic similarity of the two groups of two microRNA-associated diseases is achieved.Finally,the semantic similarity of the two groups of diseases is converted to functional similarity of microRNA pair,and we construct microRNA functional network.In addition,the effeciency of the method is inferred by microRNA in the same family,microRNA in the same cluster and that of microRNA that belong to neither the same family nor the same cluster.MFSS is proved that it has higher performance than the existing microRNA function similarity calculation method for 15 kinds of human diseases.(2)On the basis of constructing microRNA functional network,a method based on random walk is proposed for predicting disease-related microRNA.We construct microRNA functional network on the basis of the accurately calculating microRNA functional similarity for converting the prediction of disease-related microRNA into random walk problem.A new prediction method,MIDP,based on random walk on the functional network is proposed in the thesis.The prior information of nodes in the microRNA functional network can be completely exploited in the method.For the diseases with some known related mi RNAs,the network nodes are divided into marked nodes that are known to be associated with a specific disease and unmarked nodes have not been found to be associated with a specific disease,and the transition matrices are established for the two categories of nodes by combining the different transition weights of the two categories of nodes on the basis of the functional similarity of microRNA pair.A random walker starts at one of known specific disease-related marked nodes with equal probability and random walk in the functional network according to the transition matrices.After the iterative walking process is converged,the steady-state probability with which the walker stays at a candidate node is defined as its relevance score between the specific disease and the microRNA node.In this way,the unmarked nodes with higher scores are more likely to be used as microRNA candidates associated with the specific disease.The efficiency of MIDP is compared with other prediction methods by the association data of 15 human diseases.The experimental result indicates that our methods can effectively predict microRNA candidates associated with specific diseases and achieve superior prediction performance.(3)On the basis of constructing microRNA-disease bilayer network,a method based on non-negative matrix factorization is proposed for predicting disease-related microRNA.Currently,disease-related microRNA prediction methods relied on microRNA that have already been related to the specific disease and therefore are not effective on the new diseases without any known related microRNA.We construct a bilayer network to represent the complex relationships among microRNA,among diseases and between microRNA and diseases,and propose a prediction method of disease-related microRNA based on non-negative matrix factorization with sparse constraints.The method is referred to as DMPred.The method integrates the microRNA functional similarity,the disease similarity,and the microRNA-disease associations seamlessly,which exploits the complex relationships within the bilayer network and the consensus relationship between multiple kinds of information.Considering the correlation between the candidates related to various diseases,it predicts their respective candidates for all the diseases simultaneously.In addition,the sparseness characteristic of disease-microRNA was introduced to generate more reliable prediction model.The results of cross validation and simulation experiments on 15 common diseases confirmed the superiority of DMPred for the specific disease-related microRNA and the new diseases without any known related microRNA.更多还原