论文部分内容阅读
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 study the Bayesian interpretation and empirical Bayesian estimation of the false discovery rate in multiple hypothesis tests. Methods Bayesian models of prostate cancer microarray data were established. Empirical methods were used to estimate the z-value distribution function and Poisson regression method to estimate the marginal density of z values. Empirically, Bayesian error detection rate and local error detection rate were estimated. Results The empirical Bayesian estimation of error detection rate (-∞, -3) as rejection domain was 0.167, and the number of genes with local error detection rate below 0.20 was 58. Conclusion The error can be explained from Bayesian statistics The discovery rate, the false discovery rate can be estimated empirically in high-dimensional data.