Tang Yongqiang, Ghosal Subhashis, Roy Anindya
Department of Psychiatry, SUNY Health Science Center, Brooklyn, New York 11203, USA.
Biometrics. 2007 Dec;63(4):1126-34. doi: 10.1111/j.1541-0420.2007.00819.x. Epub 2007 May 14.
We propose a Dirichlet process mixture model (DPMM) for the P-value distribution in a multiple testing problem. The DPMM allows us to obtain posterior estimates of quantities such as the proportion of true null hypothesis and the probability of rejection of a single hypothesis. We describe a Markov chain Monte Carlo algorithm for computing the posterior and the posterior estimates. We propose an estimator of the positive false discovery rate based on these posterior estimates and investigate the performance of the proposed estimator via simulation. We also apply our methodology to analyze a leukemia data set.
我们针对多重检验问题中的P值分布提出了一种狄利克雷过程混合模型（DPMM）。该DPMM使我们能够获得诸如真原假设比例和单个假设被拒绝概率等数量的后验估计。我们描述了一种用于计算后验和后验估计的马尔可夫链蒙特卡罗算法。我们基于这些后验估计提出了一种正错误发现率的估计器，并通过模拟研究了所提出估计器的性能。我们还应用我们的方法来分析一个白血病数据集。
