Powerful and interpretable control of false discoveries in two-group differential expression studies

Area of Science:

• Genomics
• Statistical Bioinformatics
• Computational Biology

Background:

• Standard differential expression (DE) analysis relies on controlling the false discovery rate (FDR), which lacks guarantees on the proportion of false discoveries.
• Existing FDR control methods offer limited statistical assurances, especially for subsets of selected genes.

Purpose of the Study:

• To demonstrate the effectiveness of adaptive interpolation-based post hoc inference methods for two-group differential expression studies.
• To address the limitations of FDR control by providing statistical guarantees on false discoveries.

Main Methods:

• Formalized permutation-based methods for sharp, gene-dependent confidence bounds.
• Developed a linear time algorithm for computing post hoc bounds, enabling large-scale DE studies.
• Introduced the Adaptive Simes bound for robust statistical inference.

Main Results:

• Adaptive post hoc methods provide reliable guarantees on the number or proportion of false discoveries.
• The Adaptive Simes bound demonstrated strong statistical performance in numerical experiments.
• Illustrated the application of the Adaptive Simes bound on a real RNA sequencing study.

Conclusions:

• Adaptive post hoc inference offers a statistically rigorous alternative to FDR control in DE analysis.
• The developed methods are computationally efficient and applicable to large-scale genomic data.
• Open-source implementation available in the R package 'sanssouci'.