基因表达数据聚类分析方法的比较  

Comparison of cluster analysis methods for gene expression profile

作　　者：张瑾[1,2] 岳超[1,3] 章元明[1] 

机构地区：[1]南京农业大学作物遗传与种质创新国家重点实验室,江苏南京210095 [2]南京农业大学理学院,江苏南京210095 [3]江苏中烟工业有限责任公司,江苏南京210019

出　　处：《南京农业大学学报》2014年第6期1-6,共6页Journal of Nanjing Agricultural University

基　　金：中央高校基本科研业务费青年基金项目(KJQN201414);国家自然科学基金青年基金项目(31301229)

摘　　要：通过生物信息学分析方法,利用广泛使用的基因芯片技术产生的数万个基因表达数据,揭示基因的功能和相互作用。聚类分析是一种主要的生物信息学分析方法,能高效发掘功能一致的基因。针对基因表达谱聚类分析方法较多、应用者选择方法困难的问题,本研究利用3组基因表达谱模拟数据和1组酵母菌基因表达实际数据,通过Caliński-Harabasz指数、灵敏值和分类正确率3个指标,比较了平滑样条聚类、数量性状关联聚类和局部逼近模糊聚类法3种经典方法。结果表明:平滑样条聚类法的Caliński-Harabasz指数平均数最大,灵敏值平均数最小,分类正确率最大,为最优方法;数量性状关联聚类次之,局部逼近模糊聚类最差。这一结果为今后基因表达谱数据聚类分析方法选择提供了参考依据。Gene expression data, derived from widely used DNA microarray technology, for tens of thousands of genes was used to unravel gene functions and interactions using bioinformatics approaches. Of these approaches, cluster analysis method is one major method, and can be used to mine similar function genes. However, it is difficuh to select one suitable method from many current approaches. To solve this issue, in this study three simulated gene expression datasets and one gene expression dataset in yeast were analyzed by three classic cluster analysis methods：smoothing spline clustering, quantitative trait associated cluster analysis and fuzzy clustering by local approximation of membership, and the performances of all the three methods were reflected by three indicators： Califiski-Harabasz index, figure of merit and correct classification rate. Results showed that maximum average Califiski-Harabasz index and correct classification rate, and minmum average figure of merit were observed from smoothing spline clustering, being the best method;quantitative trait associated cluster analysis followed smoothing spline clustering;and fuzzy clustering was the worst. This finding provides a good foundation in the selection of cluster analysis approaches.

关 键 词：聚类分析 Califiski-Harabasz指数 灵敏值 分类正确率 基因表达谱 

分 类 号：Q348[生物学—遗传学]