False discovery rate estimation for large‐scale homogeneous discrete p ‐values

Summary Large‐scale homogeneous discrete p ‐values are encountered frequently in high‐throughput genomics studies, and the related multiple testing problems become challenging because most existing methods for the false discovery rate (FDR) assume continuous p ‐values. In this article, we study the estimation of the null proportion and FDR for discrete p ‐values with common support. In the finite sample setting, we propose a novel class of conservative FDR estimators. Furthermore, we show that a broad class of FDR estimators is simultaneously conservative over all support points under some weak dependence condition in the asymptotic setting. We further demonstrate the significant improvement of a newly proposed method over existing methods through simulation studies and a case study.