贝叶斯错误发现率  

Bayesian false discovery rate

作　　者：曾平[1] 王婷[1] 

机构地区：[1]徐州医学院流行病与卫生统计学教研室,江苏徐州221002

出　　处：《山东大学学报（医学版）》2012年第3期120-124,共5页Journal of Shandong University:Health Sciences

基　　金：江苏省教育厅高校哲学社会科学研究基金资助项目(2010SJB790037)

摘　　要：目的研究多重假设检验中错误发现率的贝叶斯解释和经验贝叶斯估计。方法建立前列腺癌微阵列数据的贝叶斯模型,采用经验的方法估计z值的分布函数和应用Poisson回归方法估计z值的边际密度,然后经验估计贝叶斯错误发现率和局部错误发现率。结果以(-∞,-3]作为拒绝域时错误发现率的经验贝叶斯估计值为0.167,局部错误发现率在0.20以下的基因有58个。结论可从贝叶斯统计的角度解释错误发现率,在高维数据中能够经验地估计错误发现率。Objective To investigate the Bayesian interpretation of false discovery rate in multiple hypothesis testing and the empirical Bayes approach.Methods A Bayesian two-group model was constructed for prostate cancer microarray data.The cumulative distribution function of the z value was empirically estimated and the density function was estimated using the method of Poisson regression by Efron,then the empirical Bayes approach was applied to estimate false discovery rate and local false discovery rate.Results The false discovery rate was empirically estimated as 0.167 given the rejection region of（-∞,-3],and the local false discovery rates were found to be no more than 0.2 in 58 genes.Conclusion The false discovery rate can be interpreted from the Bayesian perspective and empirically estimated in high-dimensional data.

关 键 词：大规模数据 多重假设检验 错误发现率 贝叶斯和经验贝叶斯 密度估计 前列腺肿瘤 

分 类 号：R195.1[医药卫生—卫生统计学]