【作者】 石鑫；

【导师】 姜琳颖； 王健；

【作者基本信息】 东北大学 ， 软件工程（专业学位）， 2016， 硕士

【摘要】 基因决定了一切生命从出生到死亡整个生命过程的表现形式。通过基因检测与数据分析可以帮助人们知道人一生的疾病走向与潜在隐患,并引导医学向更准确的治疗方向发展。基因表达数据是通过微阵列技术得到,即DNA微阵列技术,又被成为基因芯片技术。通过基因微阵列技术,人们可以获得大量的基因表达数据。对有着生物信息的癌症基因数据进行分析与研究,将有助于疾病的预测与诊断。然而,如何对有着数据维数高、样本小、相关性大、冗余特征和噪声特征干扰性强等特点的基因微阵列数据进行快速、高效、准确的提取与分类,已经成为当前针对基因微阵列数据研究的重要课题之一。本文在阅读大量国内外相关文献,多次实验的基础上,以癌症基因微阵列数据为研究对象,开发了基因微阵列数据分类系统。本文由基因微阵列数据的特征选择、特征提取、分类等部分组成。在特征选择与特征提取部分,利用小波变换将其快速的降维,减少冗余特征数据对样本分类的影响。鉴于微阵列实验复杂性与环境等因素的影响,以及采集数据对象又存在差异,这使得相同分类器在不同数据集上,表现出不同的学习效果。文本将通过改进Borda融合算法,将多特征选择方法得到的特征排序序列,融合成一个最优的特征排序结果。同时,还研究了用粒子群优化算法对小波系数进行筛选,进而得到最优的特征子集。为了验证本文算法能够有效地实现基因数据分类,并且其系统具有一定的可行性与实用性。本文将在三个数据集上进行实验,分别是急性白血病数据集、前列腺数据集、肺癌数据集。实验结果表明:本方法能得到较好的微阵列数据分类结果。并且将改进后的算法应用到分类系统中,不仅进一步验证了本文算法的有效性,同时验证了基因微阵列数据分类系统的合理性与可用性。更多还原

【Abstract】 Genes determine all manifestation of life from birth to death throughout life.Through genetic testing and data analysis can help people know the direction of human diseases and potential pitfalls,and guide medicine to a more accurate direction of treatment.Gene expression data is obtained by microarray technology,DNA microarray technology,has also been a gene chip technology.Through gene microarray technology,people can get a lot of gene expression data.The analysis and research of cancer gene data with biological information will help to predict and diagnose the disease.However,how to have the data of high dimension,small samples,correlation,redundancy and noise characteristics of interference characteristics of the gene microarray data for rapid,efficient and accurate extraction and classification,has become one of the important topics in the study of gene microarray data.In this paper,based on reading a lot of domestic and foreign literature,a number of experiments on the basis of cancer gene microarray data as the research object,developed a gene microarray data classification system.This paper is composed of feature selection,feature extraction and classification of gene microarray data.In the part of feature selection and feature extraction,the wavelet transform is used to reduce the dimension quickly and reduce the influence of redundant feature data on sample classification.In view of the complexity of microarray experiments and the influence of environmental factors,and the differences of data acquisition objects,this makes the same classifier show different learning effects on different data sets.Text will be improved by Borda fusion algorithm,the feature selection method to obtain the characteristics of the sort sequence,fused into an optimal feature sequencing results.At the same time,the particle swarm optimization(PSO)algorithm is used to filter the wavelet coefficients to obtain the optimal feature subset.In order to verify the algorithm can effectively achieve the classification of genetic data,and its system has a certain feasibility and practicality.In this paper,we will carry out experiments on three data sets,namely acute leukemia data sets,prostate data sets,lung cancer data sets.The experimental results show that this method can get better results of microarray data classification.Furthermore,the improved algorithm is applied to the classification system,which not only validates the validity of this algorithm,but also verifies the rationality and usability of the gene microarray data classification system.更多还原