LoG矩阵分解的肿瘤基因特征提取方法

伴随着基因芯片的发展,通过研究海量的基因表达谱数据来识别肿瘤已成为生物信息学研究的热点.提出一种基于LoG(Laplace of Gaussian)矩阵分解的肿瘤基因特征提取方法,该方法首先将样本数据映射为高维空间中的点,然后构建点与点之间的LoG矩阵,在保留样本分类信息的情况下,使得无结构信息的基因表达谱数据变成具有结构信息的图,再对LoG权值矩阵进行非负矩阵分解得到能够表征样本特征的特征分量,最后用KNN对样本进行分类.通过对白血病和结肠癌基因表达谱数据的特征提取,验证该文方法的可行性和有效性.

Tumor feature extraction based on LoG matrix factorization

With the development of gene microarray,how to analyze the tumor gene expression profiles to recognize the types of tumor has become the focus of bioinformation.A new algorithm based on LoG matrix factorization was proposed to extract the feature of gene expression profiles.Firstly,the samples were mapped into the high-dimension space,and then the LoG matrix which contains the information of all samples was constructed.In this way,the gene expression profiles with unstructured information could be transformed into the structured graph.Then the LoG matrix was factorized by NMF and got the eigenvectors which could represent the feature of samples.Finally,classification experiments were implemented by KNN.The feasibility and effectiveness of this algorithm were verified through the tests on leukemia data and colon data.