Title:DART2: a robust multiple testing method to smartly leverage helpful or misleading ancillary information

Abstract:In many applications of multiple testing, ancillary information is available, reflecting the hypothesis null or alternative status. Several methods have been developed to leverage this ancillary information to enhance testing power, typically requiring the ancillary information is helpful enough to ensure favorable performance. In this paper, we develop a robust and effective distance-assisted multiple testing procedure named DART2, designed to be powerful and robust regardless of the quality of ancillary information. When the ancillary information is helpful, DART2 can asymptotically control FDR while improving power; otherwise, DART2 can still control FDR and maintain power at least as high as ignoring the ancillary information. We demonstrated DART2's superior performance compared to existing methods through numerical studies under various settings. In addition, DART2 has been applied to a gene association study where we have shown its superior accuracy and robustness under two different types of ancillary information.