β散度的非负矩阵分解在基因聚类中的应用.PDFVIP

第8 卷 第4 期 Vol.8 No.4 2015 年 2 月 February 2015 β 散度的非负矩阵分解在基因聚类中的应用 崔 建，何光辉* （重庆大学数学与统计学院，重庆 401331 ） 摘要：针对传统的非负矩阵分解（non-negative matrix factorization ，NMF ）算法在基因表达数据聚类应用 中的低效性，研究采用一种基于β 散度的 NMF （β-NMF ）算法，该方法中β 的选择是一个开放性问题。实验 中β 取不同值分别对基因表达数据进行分解，再通过 K 均值聚类。分析结果并与传统的基于梯度下降的 NMF 和 Kullback–Leibler （KL ）散度的NMF 实验做对比，得出当 β 取值为 0.5 时，该算法对基因数据表达谱具有 较好的聚类效果。 关键词：模式识别；β 散度；非负矩阵分解；基因表达数据；聚类；梯度下降 中图分类号：TP391.19 文献标识码：A 文章编号：1674-2850(2015)04-0325-06 Application of β-divergence non-negative matrix factorization in gene clustering CUI Jian, HE Guanghui (College of Mathematics and Statistics, Chongqing University, Chongqing 401331, China) Abstract: Traditionally non-negative matrix factorization (NMF) algorithm clustering for the application of gene expression data is inefficient. In this paper, we used NMF algorithm based on β divergence (β-NMF) to avoid some insufficiency of classical NMF algorithm. It was an open problem to choose the best optimal value of β. In our experiments, we took different value of β for gene expression data decomposition, and then clustered gene expression data by K-means algorithm. By analyzing and comparing experimental results with traditional non-negative matrix decomposition based on gradient descent and Kullback–Leibler (KL) divergence decomposition, we concluded that when β was 0.5, the algorithm had an better clustering effect. Key words: pattern reco