A0910
Title: FDR control for high dimensional quantile variable selection
Authors: Tianhai Zu - University of Texas at San Antonio (United States) [presenting]
Zhigen Zhao - Temple University (United States)
Yan Yu - University of Cincinnati (United States)
Abstract: Multiple testing is a significant challenge in genetic research, especially when investigating complex diseases. Quantile regression is increasingly critical for providing a nuanced understanding of heterogeneous relationships between genetic markers and complex conditions like diabetes. However, existing mechanisms for false discovery rate (FDR) control are not tailored to the quantile regression framework. To tackle these challenges, a novel FDR control method is proposed for linear quantile regression, utilizing data-splitting mirror statistics. The approach addresses current limitations in existing FDR control methods for quantile regression and is especially advantageous in preserving high power. Theoretical justifications are provided, highlighting that this is the first attempt for controlling FDR in linear quantile regression. Extensive simulations confirm the efficacy of the approach. Furthermore, its use case is demonstrated through a case study on diabetes data, with particular emphasis on high-risk quantiles. The method effectively identifies genetic factors across various diabetes risk quantiles that may benefit improved diagnostics and treatments.
