How accurately can we control the FDR in analyzing microarray data?

Area of Science:

• Bioinformatics
• Statistical Genetics
• Genomics

Background:

• Microarray data analysis often uses False Discovery Rate (FDR)-based multiple testing procedures.
• Standard FDR procedures assume independence or weak dependence of test statistics, which is often violated in gene expression data due to biological and technical correlations.
• High dimensionality of genomic data makes verifying these assumptions challenging.

Purpose of the Study:

• To evaluate the performance of Benjamini-Hochberg and Storey FDR procedures on real microarray data.
• To propose and validate a simulation-based method for assessing FDR control without assuming test statistic independence.
• To directly examine the practical performance of FDR procedures in the context of correlated gene expression data.

Main Methods:

• Generating simulated test statistics that mimic the asymptotic correlation structure of real microarray data.
• Utilizing simulation models to directly assess the FDR control of Benjamini-Hochberg and Storey procedures.
• Applying the proposed simulation method to real microarray datasets with disease group and survival endpoints.

Main Results:

• The study demonstrates a method to evaluate FDR procedures on real data by simulating correlated test statistics.
• Performance of FDR procedures is assessed under realistic correlation structures found in gene expression data.
• The approach provides a direct assessment of FDR control, circumventing the need to verify theoretical independence assumptions.

Conclusions:

• The proposed simulation-based approach offers a practical way to evaluate FDR-based multiple testing procedures for genomic data.
• This method allows for a direct assessment of FDR control, even when test statistics exhibit complex correlation patterns.
• The findings are relevant for accurate statistical inference in high-dimensional genomic studies, including those involving disease and survival analysis.