基于多正则约束非负矩阵分解的基因特征提取

摘要

针对基因表达谱数据高维度、高噪声的特点,在传统非负矩阵分解(NMF)理论的基础上,提出一种基于多正则约束非负矩阵分解(MRCNMF)的特征提取模型.通过引入流形正则,使得NMF在维数约简的同时能够保持原始数据的内部空间结构,低秩稀疏正则约束对噪声和数据丢失具有较好的抑制作用.提出一种模型求解方法,通过引入+乘子保持矩阵分解的非负性.实验结果表明,采用的特征提取算法对基因表达谱中的噪声具有较强的抑制作用,与NMF和图正则非负矩阵分解(GNMF)相比能够达到更高的分类精度.%Aiming at the problems of gene expression profiling,such as high dimension and noise,a feature extraction model based on multi-regularized constraints non-negative matrix factorization (MRCNMF) was presented based on the traditional non-negative matrix factorization (NMF).By introducing manifold regularized,the NMF kept internal spatial structure of the original data effectively and reduced dimension at the same time.Low-rank sparse constraints had good inhibition to noise and data loss.A method that kept the non-negativity of matrix by introducing multiplier + was proposed to solve the model.Experimental results show that the feature extraction algorithm has strong inhibition to the noise of gene expression profiling.Compared with NMF and graph regularized non-negative matrix factorization (GNMF),it can achieve higher clustering accuracy.