Abstract
Abstract
The symmetrized data aggregation (SDA) method has obvious advantages in controlling false discovery rate (FDR) under dependence, but it is affected by data sparsity level, signal amplitude and feature correlation structures. The FDR control becomes less accurate due to the additional estimation errors. In this paper, we expand the symmetrized data aggregation (SDA) filter by using sorted L-one penalized estimation (SLOPE) method, and in comparison with SDA, SDA-SLOPE method can estimate FDR more accurately and improve the true discovery rate (TDR). Through simulation study, it is found that SDA-SLOPE can adapt to the changes of data sparsity, signal amplitude and feature correlation, and the robustness and effectiveness of this method in FDR control are verified.
Subject
General Physics and Astronomy
Reference12 articles.
1. Controlling the false discovery rate: a practical and powerful approach to multiple testing;Benjamini;Journal of the Royal Statistical Society,1995
2. Multiple hypotheses testing with weights;Benjamini;Scandinavian Journal of Statistics,1997
3. Estimating false discovery proportion under arbitrary covariance dependence;Fan;Journal of the American statal association,2012
4. Estimation of the false discovery proportion with unknown dependence;Fan;Journal of the Royal Statistical Society: Series B,2017
5. A rate optimal procedure for recovering sparse differences between high-dimensional means under dependence;Li;The Annals of Statistics,2017