Adaptive Filtering Multiple Testing Procedures for Partial Conjunction Hypotheses
主 题: Adaptive Filtering Multiple Testing Procedures for Partial Conjunction Hypotheses
报告人: Jingshu Wang ( University of Pennsylvania)
时 间: 2016-12-30 14:00 - 15:00
地 点: 理科一号楼1418
The partial conjunction (PC) alternative hypothesis $H_1^{r/n}$ stipulates that at least r of n related basic hypotheses are non-null, making it a useful measure of replicability. Motivated by genomic problems we consider a setting with a large number M of partial conjunction null hypotheses to test, based on an $n\times M$ matrix of p-values. When r > 1 the hypothesis $H_0^{r/n}$ is composite. Validity versus the case with r ? 1 alternative hypotheses holding can lead to very conservative tests. We develop a filtering approach for $H_0^{r/n}$ based on the M p-values for $H_0^{(r-1)/n}$. This filtering approach has greater power than straightforward PC testing in the multiple testing setting. We prove that it can be used to control the familywise error rate, the per family error rate, and the false discovery rate among M PC tests. In simulations we find that our filtering approach properly controls the FDR while achieving good power. We illustrate application of the method in both microarray data analysis and genome-wide association studies (GWAS). About the speaker: Dr. Jingshu Wang received B.S. from Peking University in 2011 and Ph.D. from Stanford University in 2016, and is currently Postdoctoral Researcher in the Department of Statistics at the Wharton School of the University of Pennsylvania. Her research has been focused on high-dimensional factor models, multiple hypotheses testing and meta-analysis, and their applications in genetics problems, and has been published in the Annals of Statistics and Statistical Science.
