微阵列基因表达数据混合特征算法研究

发布时间：2018-03-06 07:04

本文选题：微阵列基因表达　切入点：Filter法　出处：《沈阳工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：DNA微阵列技术是由生物学,融微电子学、计算机科学和广电化学为一体,在原来核酸杂交的基础上发展起来的一项新技术,在医学和生物学的研究中得到了高度重视。近几年来,随着大规模高通量微阵列技术的快速发展,在一次实验中获得成千上万个基因的表达水平成为现实。这种高新的技术为基因表达数据的搜集提供了便利,从一次实验获得大量的反应基因产物mRNA丰度的数据,通过微阵列技术得到的反应mRNA丰度的数据通常称为微阵列基因表达数据,简称微阵列。自上个世纪90年代以来微阵列技术逐步形成,对生物领域产生深远影响,它的出现使基因活性的检测成为可能,将微阵列技术应用病理诊断分析的实验便由此开始。此后,在经历了20多年的不断发展,将生物医学,计算机等领域将其融合。如今,微阵列技术不断完善,成为生物信息学中热门话题,为人类探索生物信息提供了新的篇章。由于微阵列的维数高,噪声大,及冗余性强等特性,这种特征给基因选择的实验带来了挑战;本文提出一种面向高维微阵列数据的混合特征算法,根据信噪比方法,Lasso方法,Filter方法及Wrapper方法混合优势对微阵列基因表达数据分析,提出混合Relief-PSO方法对数据分析研究。本文首先回顾微阵列数据分析的一些方法,然后介绍集成系统学习的方法。最后是关于新的集成技术在微阵列数据上的应用。主要工作如下:(1)微阵列算法的介绍,以及对于单独的信噪比方法,Lasso方法,Filter方法,Wrapper方法的特征在微阵列的应用。借于结合的信噪比和Lasso法、Filter法和Wrapper法混合。最后阐明本文的主要研究内容。(2)单个实验产生的微阵列数据难免会达不到理想状态,也会影响到数据学习训练分类器的效果和泛化能力,将多个实验的方法集成起来可以提高分类器泛化能力和实验数据的高效性,也更加接近基因表达数据的分析能力,在处理高维小样本、高冗余、高噪声的基因微阵列数据时,无法采用传统特征选择方法进行分析。本文针对该问题提出了一种结合Relief算法和粒子群优化算法(Relief-PSO)的混合特征选择方法,检测这种算法对数据的分类效果,产生泛化能力更好的集成算法。
[Abstract]:DNA microarray technology is a new technology developed by combining biology, microelectronics, computer science and radio and television chemistry, based on the original nucleic acid hybridization. In recent years, with the rapid development of large-scale high-throughput microarray technology, In one experiment, the expression level of tens of thousands of genes became a reality. This new technique facilitated the collection of gene expression data, and obtained a large amount of mRNA abundance data from one experiment. The mRNA abundance response data obtained by microarray technology is commonly called microarray gene expression data, or microarray. Since -10s, microarray technology has been gradually formed, which has a profound impact on the biological field. It made it possible to detect gene activity, and the experiment of applying microarray technology to pathological diagnosis and analysis began. Then, after more than 20 years of continuous development, it fused biomedical, computer, and other fields. Microarray technology has become a hot topic in bioinformatics, which provides a new chapter for human beings to explore bioinformatics. In this paper, a hybrid feature algorithm for high-dimensional microarray data is proposed to analyze the gene expression data of microarrays according to the mixed advantages of the signal-to-noise ratio (SNR) method, Lasso method, filter method and Wrapper method. A hybrid Relief-PSO method for data analysis is proposed. Firstly, some methods of microarray data analysis are reviewed. Finally, the application of new integration technology in microarray data is introduced. The main work is as follows: introduction of microarray algorithm. And the application of the characteristics of the Lasso / filter method / wrapper method to the microarray. By combining the SNR method with the Lasso filter method and the Wrapper method. Finally, the main research content of this paper is elucidated. 2) A single experimental production. The raw microarray data will inevitably fail to reach the desired state, It can also affect the effect and generalization ability of data learning training classifier. Integrating several experimental methods can improve the generalization ability of classifier and the efficiency of experimental data, and it is also closer to the analysis ability of gene expression data. When processing high-dimensional, small-sample, high-redundancy, high-noise gene microarray data, Traditional feature selection method can not be used to analyze this problem. A hybrid feature selection method combining Relief algorithm and particle swarm optimization (PSO) algorithm is proposed in this paper to detect the effect of this algorithm on data classification. A better integration algorithm with better generalization ability is generated.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【相似文献】