Null-free False Discovery Rate Control Using Decoy Permutations.

Area of Science:

• Statistics
• Genomics
• Proteomics
• Bioinformatics

Background:

• Traditional false discovery rate (FDR) control methods rely on null distributions, which can have inherent drawbacks.
• Theoretical, permutation-based, and empirical null distributions may fail due to improper assumptions or limitations.
• Existing methods struggle with complex dependencies often found in real-world biological data.

Purpose of the Study:

• To propose a novel, null distribution-free approach for FDR control in multiple hypothesis testing.
• To introduce the 'target-decoy procedure' for robust FDR control in case-control studies.
• To demonstrate the stability and power of the new method compared to existing techniques.

Main Methods:

• Developed a 'target-decoy procedure' that does not require knowledge of the null distribution.
• Utilized competitive decoy tests, generated from sample permutations, to estimate false discoveries.
• Established theoretical guarantees for FDR control under score symmetry and independence assumptions.

Main Results:

• The target-decoy procedure effectively controls FDR without relying on null distribution assumptions.
• Simulations show the proposed method is more stable and powerful than traditional FDR control approaches.
• The procedure performs well even in the presence of data dependency, outperforming existing methods.

Conclusions:

• The target-decoy procedure offers a robust and powerful alternative for FDR control in multiple hypothesis testing.
• This null distribution-free approach is particularly valuable for complex biological datasets, such as genomics and proteomics.
• The method demonstrated superior performance on both simulated and real-world datasets, including Arabidopsis and COVID-19 data.