差异甲基化位点识别算法的对比研究与优化

发布时间：2018-04-10 18:25

本文选题：DNA甲基化 + 差异甲基化位点　；参考：《电子科技大学》2017年硕士论文

【摘要】：DNA甲基化是最重要的表观遗传机制之一,在基因表达调控、胚胎发育、X染色体失活、基因印记以及维持染色质结构等生物学过程中发挥着重要的作用,并且与许多疾病的致病过程相关。近年来的大量研究表明,差异甲基化位点与许多疾病有直接关系,特别是癌症。因此,识别差异甲基化位点是解剖疾病病因中最关键和最根本的问题之一。本文对多种癌症进行分析,找到每种癌症对应的差异甲基化位点,为癌症早期诊断提供依据,为此本文做了如下工作:1)针对现有统计假设检验方法只是选出在统计上有显著差异的位点,选出来的位点并不都具有类别区分特性,本文引入了基于机器学习的特征选择方法(Elastic Net正则化)。该方法有效解决了假设检验方法不能同时发现多个位点对癌症的组合作用。2)针对在优化差异甲基化识别算法过程中发现选择出的差异甲基化位点不稳定问题,本文提出了基于Elastic Net正则化的集成特征选择算法。本文选择了 13种癌症数据来分析算法特征选择稳固性,结果发现在两种特征选择算法模型分类性能接近的情况下本文算法在特征选择稳固性评价指标(杰卡德指数)上优于Elastic Net正则化特征选择算法。3)在与现有统计假设检验方法的对比中,本文采用在独立测试集上测试本文算法选择出的差异甲基化位点与FastDMA、RnBeads两种假设检验方法得到的差异甲基化位点的类别区分性能。结果发现,本文算法在独立测试集上正确率高于FastDMA与RnBeads,由此可知本文算法选出的差异甲基化位点的类别区分性能优于两种假设检验方法。4)针对本文选出的差异甲基化位点是否有实际的生物意义,本文采用了对多种癌症做癌症共性分析。本文将癌症共有差异甲基化位点对应到基因上,得到38个共有差异基因,查阅文献发现有23个共有差异基因跟癌症有密切关系。本文进行pathway分析时,发现共有代谢通路有11个,其中9个和癌症有直接关系。这说明本文找到的差异甲基化位点能为生物学家提供有效参考。
[Abstract]:DNA methylation is one of the most important epigenetic mechanisms. It plays an important role in gene expression regulation, embryo development X chromosome inactivation, gene imprinting and maintaining chromatin structure.And related to the pathogenesis of many diseases.A large number of studies in recent years have shown that differential methylation sites are directly related to many diseases, especially cancer.Therefore, the identification of differential methylation sites is one of the most critical and fundamental problems in the etiology of anatomical diseases.In this paper, the differential methylation sites for each cancer were found by analyzing various kinds of cancers, which provided the basis for the early diagnosis of cancer.In this paper, the following work has been done: (1) in view of the existing statistical hypothesis testing methods, only the sites with significant statistical differences are selected, and not all of the selected loci have the characteristics of category differentiation.In this paper, an Elastic Net regularization method based on machine learning is introduced.This method effectively solves the problem that the hypothesis test method can not find the combination action of multiple sites on cancer at the same time. 2) in the process of optimizing the differential methylation identification algorithm, we can find the instability of the selected differential methylation sites.This paper presents an ensemble feature selection algorithm based on Elastic Net regularization.In this paper, 13 kinds of cancer data are selected to analyze the stability of algorithm feature selection.The results show that the proposed algorithm is better than Elastic Net regularization feature selection algorithm. 3 on the evaluation index of feature selection stability (Jekard exponent) under the condition that the classification performance of the two feature selection algorithms is similar to that of the existing statistical false.In the comparison of test methods,In this paper, the differential methylation sites selected by our algorithm and the differential methylation sites obtained by FastDMA-RnBeads are tested on an independent test set.And it turns out,The accuracy of this algorithm on the independent test set is higher than that of FastDMA and RnBeads.The classification performance of the differential methylation sites selected by this algorithm is better than that of the two hypothetical test methods .4The differential methylation sites selected in this paper are as follows:Whether it has practical biological significance,In this paper, the cancer commonness analysis of many kinds of cancer is adopted.In this paper, 38 common differential genes were obtained by mapping the common differential methylation sites of cancer to genes, and 23 common differential genes were found to be closely related to cancer.Pathway analysis showed that there were 11 metabolic pathways, 9 of which were directly related to cancer.This indicates that the differential methylation sites found in this paper can provide effective reference for biologists.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R73-3

【相似文献】