Abstract: ICA is a statistical method of blind source separation.The gene linear profile model，composed of model profiles and coefficients，is obtained by ICA from gene expression data，so gene classification based on ICA is presented.But a model profile is not consistent with an average profile of a group genes completely，and model profiles are not orthogonal，so the linear profile model can’t express the gene data effectively.This paper proposes the mode space based on ICA，restructure the gene expression data，and introduce a classification method in the space.Experiments show that this method has better performance in gene classification than the method in the linear profile model of genes.

Key words: Independent Components Analysis（ICA）, gene classification, mode space, linear profile model

摘要： ICA是应用于盲源信号分离的一种统计方法。利用ICA对基因微阵列表达谱数据进行分解获得由基因模型谱和对应系数构成的线性谱模型，并在此基础上进行基因分类。由于基于ICA的一个模型谱并不能完整地代表一个具有生物意义的类别，并且模型谱之间不具正交性，在此线性模型下不能有效的表示基因数据，为此提出基于ICA的模式表达空间的概念，并在该模式空间中重新构造了基因的数据表达形式，并利用此表达形式进行基因分类。实验结果表明，该分类方法比线性谱模型下的基因分类具有更高的正确率。

关键词: 独立分量分析, 基因分类, 模式空间, 线性谱模型