小样本情况下差异表达基因鉴别的参数统计分析  

Identifying Differential Expression Genes in Small Sample Size with Parameter Statistical Methods

作　　者：贺宪民 [1] 武建虎 [1] 贺佳 [1] XIANG Zhaoying  

机构地区：[1]第二军医大学卫生统计学教研室,200433 [2]Department of Microbiology & Immunology, Weill Medical College of Cornell University, NEW YORK, USA

出　　处：《中国卫生统计》2005年第3期141-145,共5页Chinese Journal of Health Statistics

基　　金：第二军医大学青年基金资助项目(2003SQ19)

摘　　要：目的探索小样本情况下基于不同理论的统计方法在鉴别差异表达基因时的性能。方法以实验资料为基础,估计残差方差的分布参数、基因的平均表达及差异表达水平,按照一定差异比例模拟理论数据,用于分析倍数法、t检验、随机方差模型、SAM及对数后验比法的性能及特征。结果随机方差模型、SAM及对数后验比法在鉴别差异表达基因的准确性上相近,均高于t检验和倍数法,t检验又稍高于倍数法。结论倍数法的性能受极端值的影响严重,t检验在基因特异性标准误较小情况下增加鉴别的假阳性率,而随机方差模型、SAM和对数后验比法由于统计量的计算建立在多基因的基础上,鉴别的准确性较高。Objective To explore the performance of different statistical methods in identifying differential expression genes in small samples.Methods Simulating theoretical distribution data on the base of actual experiments,which include distribution parameters of residual variance,the average levels of gene expression and differential expression,and then the data is used to compare the performance of the methods such as fold change,t test,random variance model (RVM),statistical analysis of microarrays (SAM),and empirical Bayes log posterior odds.Results Inverse-Gamma distribution models true variance structure of microarray data well.The accuracies of RVM,SAM,and empirical Bayes log posterior odds in identifying differential expression genes are nearly the same,and are higher than those of t test and fold change.Conclusion Performance of fold change is influenced by the outliers,and the false positive rate of t test is higher in small samples for the proportion of genes with very small residual variances; the estimated statistics of RVM,SAM,and empirical Bayes log posterior odds are based on multiple genes,so they always have higher performance and more complex computation.

关 键 词：小样本 差异表达基因 鉴别 参数 统计分析 对数后验比法 SAM 

分 类 号：R195[医药卫生—卫生统计学]