A weighted FDR procedure under discrete and heterogeneous null distributions

Area of Science:

• Biostatistics
• Statistical Inference
• Computational Biology

Background:

• Multiple testing (MT) is crucial in analyzing large datasets, particularly in fields like genomics and drug safety.
• Existing False Discovery Rate (FDR) control methods struggle with discrete and heterogeneous p-value distributions, leading to reduced power and unreliable results.
• There is a need for FDR procedures that are group-aware, data-adaptive, and non-asymptotically conservative for discrete data.

Purpose of the Study:

• To develop and validate a novel weighted p-value-based FDR procedure (wFDR) for multiple testing in the discrete paradigm.
• To address the limitations of existing methods in handling discrete and heterogeneous p-value distributions.
• To enhance statistical power and reliability of inference in discrete data analyses.

Main Methods:

• Introduction of the weighted FDR (wFDR) procedure, which adapts to heterogeneity and discreteness of p-value distributions.
• Theoretical justification of the non-asymptotic conservativeness of the wFDR procedure under independence.
• Simulation studies comparing wFDR with six existing procedures using p-values from binomial and Fisher's exact tests.

Main Results:

• The wFDR procedure demonstrates superior power compared to six other methods in simulations for discrete p-values.
• Theoretical analysis confirms the non-asymptotic conservativeness of the wFDR procedure.
• Application to drug safety and differential methylation studies showed wFDR identified more discoveries than two existing methods.

Conclusions:

• The proposed wFDR procedure is an effective tool for multiple testing in the discrete paradigm, offering improved power and reliable inference.
• wFDR efficiently adapts to both heterogeneity and discreteness, outperforming existing methods in various scenarios.
• The wFDR procedure provides a valuable advancement for analyzing discrete data in biomedical research.