Rank-invariant resampling based estimation of false discovery rate for analysis of small sample microarray data

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Statistical significance evaluation is crucial for identifying differentially expressed genes in microarray studies.
• Traditional p-value adjustment methods (e.g., family-wise error rate) are often too conservative for large-scale microarray data.
• False Discovery Rate (FDR) is a preferred alternative for controlling false positives, but existing estimation methods may lack reliability with few replicates.

Purpose of the Study:

• To propose and evaluate a novel rank-invariant resampling (RIR) based approach for False Discovery Rate (FDR) estimation in microarray analysis.
• To address the limitations of existing FDR estimation methods when applied to datasets with a small number of replicates.

Main Methods:

• Development of a rank-invariant resampling (RIR) approach to generate a biologically relevant null distribution.
• Comparison of the RIR method's performance against four other popular FDR estimation techniques.
• Validation using both simulated and real-world microarray datasets.

Main Results:

• The proposed RIR approach demonstrated superior performance in FDR estimation compared to four other methods.
• The RIR method's null distribution maintained similar variability to observed microarray data.
• The RIR approach provided more accurate FDR estimation than SAM's random shuffling, SPLOSH, and two theoretical methods.

Conclusions:

• The RIR method offers a more accurate and reliable approach to FDR estimation in microarray studies, particularly when dealing with limited replicates.
• Existing methods like SAM and SPLOSH were found to be too liberal, while theoretical methods were too conservative.
• The RIR approach effectively balances accuracy and reliability in FDR control for genomic data analysis.