乳癌病人临床数据的降维处理及生存预测分析

发布时间：2018-09-05 15:16

【摘要】：随着基因技术的发展,可以容易地获取到癌症病人的mi RNA表达水平。近些年的研究发现,mi RNA在癌症转移的过程中扮演着重要角色。从病人的mi RNA表达水平分析病人的病情开始受到大家的关注。本文主要研究高维mi RNA表达水平数据的降维处理和结合临床数据的生存预测分析。病人的临床数据中含有大量的删失数据,进行生存分布函数求解的时候就不能忽视删失数据。本文对于删失数据的处理分别采用乘积限估计法和极大似然估计法得到乳癌病人的生存分布函数,对极大似然估计法解的存在性条件给出了证明。提出同分布插值方法对删失数据进行插值,并对同分布插值方法的误差进行分析。病人临床数据的另一大特点是高维。高维数据的处理方法可以归纳为两类:特征选取和特征抽取。本文对比了最小网格聚类、主成分分析、Isomap三种降维方法。发现Isomap方法中的最短路径预示着癌症的病情发展路径。并对最小网格聚类方法提出了去中心化的修改。最后,在降维数据的基础上,结合临床数据中的癌症不同阶段的分期数据,对比支持向量积方法和决策树方法得到的乳癌分期分类结果。通过对mi RNA表达水平数据和临床数据进行Kruskal-Wallis检验,找出与癌症分期具有相关性的miRNA。通过逐步多元回归方法建立mi RNA表达水平与病人生存期的回归关系,发现逐步回归模型选取的部分mi RNA已经被证明与癌症具有相关性。例如:hsa-mir-548t,hsa-mir-190,hsa-mir-200b等。
[Abstract]:With the development of gene technology, the expression level of mi RNA in cancer patients can be easily obtained. Recent studies have found that RNA plays an important role in cancer metastasis. From the patient's mi RNA expression level analysis of the patient's condition began to attract attention. This paper focuses on dimensionality reduction of high dimensional mi RNA expression level data and survival prediction analysis combined with clinical data. There are a lot of censored data in the patient's clinical data, so we can not ignore the censored data when solving the survival distribution function. In this paper, we obtain the survival distribution function of breast cancer patients by product limit estimation and maximum likelihood estimation, respectively, and prove the existence condition of the solution of the maximum likelihood estimation method. The same distribution interpolation method is proposed to interpolate censored data and the error of the same distribution interpolation method is analyzed. Another characteristic of patient clinical data is high dimension. The processing methods of high-dimensional data can be divided into two categories: feature selection and feature extraction. In this paper, three dimensionality reduction methods, minimum grid clustering and principal component analysis (PCA), are compared. The shortest path in Isomap method is found to predict the progression of cancer. A decentralization modification is proposed for the minimum mesh clustering method. Finally, based on the dimensionality reduction data and the staging data of different stages of cancer in clinical data, the classification results of breast cancer stages obtained by support vector product method and decision tree method are compared. Through the Kruskal-Wallis test of mi RNA expression level data and clinical data, we find out the miRNA. which has correlation with cancer stage. The regression relationship between mi RNA expression level and patient survival was established by stepwise multivariate regression method. It was found that part of mi RNA selected by stepwise regression model had been proved to be correlated with cancer. For example: hsa-mir-548t, hsa-mir-190, hsa-mir-200b, etc.
【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】：2016
【分类号】：R737.9

【相似文献】