Modified FDR controlling procedure for multi-stage analyses.

Area of Science:

• Genomics
• Statistical genetics
• Bioinformatics

Background:

• Genomic analyses, particularly microarray experiments, involve testing numerous hypotheses simultaneously.
• Initial variable set truncation using univariate regression is common but can compromise subsequent multiple testing procedures.
• Existing methods may fail to control Type I error rates after multi-stage analyses.

Purpose of the Study:

• To propose a modified Benjamini & Hochberg procedure for multi-stage analyses (FDR-MSA).
• To ensure accurate control of Type I error rates for the entire variable set in staged analyses.
• To adapt FDR control for variable importance applications in genomics.

Main Methods:

• Development of a modified marginal Benjamini & Hochberg step-up False Discovery Rate (FDR) controlling procedure for multi-stage analyses (FDR-MSA).
• Theoretical analysis showing convergence to standard Benjamini & Hochberg procedures as the initial subset size increases.
• Validation through simulation studies and application to real-world genomic data.

Main Results:

• FDR-MSA correctly controls Type I error rates in multi-stage analyses, even after initial variable subsetting.
• The proposed method demonstrates robust performance in simulations, maintaining desired error control.
• Application to the Golub Leukemia data validates the method's practical utility.

Conclusions:

• FDR-MSA provides a reliable solution for controlling false discovery rates in complex, multi-stage genomic studies.
• The method ensures statistical rigor when initial variable selection is performed before comprehensive testing.
• This approach enhances the validity of findings from microarray experiments and similar high-dimensional data analyses.