Liu Yaowu
Joint Lab of Data Science and Business Intelligence, Southwestern University of Finance and Economics, Chengdu, 611130, China.
Center of Statistical Research, Southwestern University of Finance and Economics, Chengdu, 611130, China.
Biometrics. 2025 Jan 7;81(1). doi: 10.1093/biomtc/ujaf011.
Large-scale mediation analysis has received increasing interest in recent years, especially in genome-wide epigenetic studies. The statistical problem in large-scale mediation analysis concerns testing composite null hypotheses in the context of large-scale multiple testing. The classical Sobel's and joint significance tests are overly conservative and therefore are underpowered in practice. In this work, we propose a testing method for large-scale composite null hypothesis testing to properly control the type I error and hence improve the testing power. Our method is simple and essentially only requires counting the number of observed test statistics in a certain region. Non-asymptotic theories are established under weak assumptions and indicate that the proposed method controls the type I error well and is powerful. Extensive simulation studies confirm our non-asymptotic theories and show that the proposed method controls the type I error in all settings and has strong power. A data analysis on DNA methylation is also presented to illustrate our method.
近年来，大规模中介分析越来越受到关注，尤其是在全基因组表观遗传学研究中。大规模中介分析中的统计问题涉及在大规模多重检验的背景下检验复合零假设。经典的索贝尔检验和联合显著性检验过于保守，因此在实际应用中功效不足。在这项工作中，我们提出了一种用于大规模复合零假设检验的方法，以适当地控制第一类错误，从而提高检验功效。我们的方法很简单，本质上只需要计算某个区域内观察到的检验统计量的数量。在弱假设下建立了非渐近理论，表明所提出的方法能很好地控制第一类错误且功效强大。广泛的模拟研究证实了我们的非渐近理论，并表明所提出的方法在所有情况下都能控制第一类错误且具有很强的功效。还给出了一项关于DNA甲基化的数据分析以说明我们的方法。
