基于改进非负矩阵分解的肿瘤基因表达谱特征提取

摘要

针对肿瘤基因表达谱的特点,提出基于低秩图正则非负矩阵分解(LGNMF)的特征提取方法,解决了NMF算法中缺少数据的全局信息,提升特征提取的有效性.该算法在NMF算法的基础上引入低秩图约束,提高了对数据局部和全局结构的描述,使得经过特征提取后的特征空间具有更强的分类能力.通过LGNMF算法对肿瘤基因表达谱数据集进行降维,获得低维特征空间,再使用KNN分类器对低维特征空间进行分类.通过与NMF、GNMF和RGNMF算法在四组标准肿瘤基因表达谱数据集进行对比,实验结果表明LGNMF算法能够有效提升分类效果.%According to the characteristics of the tumor gene expression profiles, we proposed a feature extraction algorithm, based on low-rank graph non-negative matrix factorization (LGNMF).It solved the lack of information on the global structure data of NMF algorithm and promoted the validity of feature extraction.The algorithm had improved the description of local and global data structures, based on NMF algorithm with low-rank graph constraints, which made feature space have stronger classification ability after feature extraction.The low-dimensional feature space was obtained by LGNMF algorithm, and it was classified by KNN classifier.We compared with the NMF, GNMF and RGNMF algorithm in four groups of standard tumor gene expression profile data sets.The experimental results show that LGNMF algorithm can improve the effect on classification.