MiRNA-疾病关联关系算法研究

发布时间：2018-05-30 22:11

本文选题：MicroRNAs + 疾病　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：MicroRNAs(mi RNAs)是一类很小的内源性非编码RNA,长度约为20-24个核苷酸,其通过碱基配对与靶m RNA的3′端非编码区相结合,导致靶mRNA的降解或翻译抑制,从而在转录后水平上调控基因表达。越来越多的研究表明,mi RNA在多种生物过程中起着重要的作用。mi RNA突变及其功能失调可能会导致各种疾病。因此,识别mi RNA与疾病之间的调控关系至关重要,进而成为近年来的一个研究热点。早期研究采用生物实验的方法,确定单一因素对实验结果的影响,获得的结果准确度较高。然而实验方法存在周期长、成本高的弊端,科研工作者致力于寻找更高效的计算方法解决该问题。因此,探究mi RNA-疾病关联关系至关重要。当前的计算方法主要分为两大类:(1)基于相似度度量的方法,(2)基于机器学习的方法。前者通过度量网络中节点之间的关联强度预测mi RNA-疾病关联,但该类方法需要构建高“质量”的生物网络模型;后者将机器学习相关算法应用到这个问题中,但这类方法需要构建高可信度的负例集合。基于上述方法存在的困难和不足,本文提出两种新的计算模型Thr RW及BNPDCMDA,用于预测mi RNAs-疾病关联关系。前者将基因引入该问题中,充分利用基因、mi RNAs和疾病自身的相似关系及三者之间的关联关系,构建非平衡的随机游走模型,预测与疾病有关联关系的mi RNAs;后者首先利用mi RNAs的功能相似度对其进行基于密度的聚类,其次结合mi RNA的聚类结果和疾病集合构建mi RNA-疾病双层子网,然后将二分网络投影应用于上述网络中,最终完成对mi RNA与疾病关联关系的预测。为了验证两种计算模型的有效性,本文选择留一交叉验证法与其它高效方法进行比较。实验结果表明:Thr RW与BNPDCMDA得到的AUC值分别可达86.24%、99.08%,明显优于当前其它方法。除此之外,本文还对某些常见疾病(如乳腺癌、肺癌)所关联的mi RNAs进行预测。实验结果表明,两种方法的预测结果均获得文献支持,从而进一步表明Thr RW与BNPDCMDA的有效性。
[Abstract]:MicroRNAs(mi RNas are a kind of small endogenous non-coding RNAs with a length of about 20-24 nucleotides. They bind to the 3 '-terminal noncoding region of the target m RNA by base pairing, leading to the degradation or translation inhibition of the target mRNA, thus regulating gene expression at the post-transcriptional level. More and more studies have shown that RNA plays an important role in many biological processes. Mi RNA mutation and its dysfunction may lead to various diseases. Therefore, it is very important to recognize the regulatory relationship between mi RNA and disease, which has become a hot topic in recent years The effect of a single factor on the experimental results was determined by biological experiments in the early stage, and the accuracy of the obtained results was high. However, experimental methods have the disadvantages of long period and high cost. Researchers are committed to finding more efficient calculation methods to solve this problem. Therefore, it is important to explore the mi RNA-disease association. The current computing methods are divided into two categories: one is based on similarity measure and the other is based on machine learning. The former predicts the miRNA-disease association by measuring the association strength between nodes in the network, but this kind of method needs to construct a "high quality" biological network model; the latter applies the machine learning correlation algorithm to this problem. However, such methods need to build a set of negative cases with high credibility. Based on the difficulties and shortcomings of the above methods, two new computational models, Thr RW and BNPDCMDA, are proposed to predict the mi RNAs-disease association. The former introduces gene into this problem, and makes full use of the similar relationship between gene RNAs and disease itself and the relationship between them to construct a non-equilibrium random walk model. The latter uses the functional similarity of mi RNAs to cluster it based on density, and then combines the clustering results of mi RNA and the disease set to construct a double layer subnet of mi RNA-disease. Then the bipartite network projection is applied to the above mentioned networks, and the prediction of the association between mi RNA and disease is finally completed. In order to verify the validity of the two computational models, this paper chooses a cross-validation method to compare with other efficient methods. The experimental results show that the AUC values obtained by the BNPDCMDA and the THR RW are 86.24 and 99.08, respectively, which are obviously superior to the other methods. In addition, we predicted the mi RNAs associated with some common diseases, such as breast cancer and lung cancer. The experimental results show that the predicted results of the two methods are supported by literature, which further demonstrates the effectiveness of Thr RW and BNPDCMDA.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R3416;TP181

【相似文献】