为从复杂的基因芯片表达数据中有效地挖掘致病基因,提出了基于奇异值分解(Singular value decomposi-tion,SVD)的致病基因挖掘算法———Logistis回归奇异值分解(Logistic regression SVD,LRSVD),针对奇异值方差评估特征模式的不足,LRSVD算法用Logistic回归系数代替方差评估每一个特征模式对分类的作用大小,进一步提出利用基因内积评估每一条基因的分类能力,建立了特征模式与原始基因表达数据之间的线性映射关系,并按基因内积大小排序,选择出对样本分类能力高的基因子集。将LRSVD算法应用于实际基因表达数据,实验结果表明,LRSVD算法能有效挖掘出与疾病相关的基因子集。 In order to effectively mine the causative genes from the complex gene expression data, a pathogenic gene mining algorithm based on singular value decomposi- tion (SVD), Logistic regression SVD (Logistic regression SVD) LRSVD) to evaluate the feature mode of singular value variance. LRSVD algorithm uses logistic regression coefficient instead of variance to evaluate the effect of each feature mode on classification, and further proposes to use gene inner product to evaluate the classification ability of each gene and establish the characteristic mode And the original gene expression data between the linear mapping relationship, and according to the size of the gene sequence, select the sample classification ability of gene subset. The LRSVD algorithm was applied to the actual gene expression data. The experimental results show that the LRSVD algorithm can effectively mine the gene subset related to disease.