Robust meta-analysis for large-scale genomic experiments based on an empirical approach

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• High-throughput technologies enable simultaneous analysis of thousands of genes.
• Genomic studies generate large datasets with significance testing results for numerous genes.
• Researchers often combine results from multiple genomic studies using meta-analysis.

Purpose of the Study:

• To address the limitations of classical meta-analysis methods in large-scale genomic studies.
• To propose a robust meta-analysis method that accounts for the large number of genes and potential hidden confounders.
• To improve the accuracy of significance testing in genomic meta-analyses.

Main Methods:

• Discusses the limitations of standard meta-analysis techniques like Fisher's p-value combination and Stouffer's Z.
• Proposes a robust meta-analysis method that empirically modifies individual test statistics and p-values prior to combination.
• Evaluates the proposed method using simulation studies and real genomic data analysis.

Main Results:

• The proposed robust meta-analysis method demonstrates superior performance in significance testing compared to classical approaches.
• The method is particularly effective in the presence of hidden confounders, reducing false discoveries.
• Simulation studies and real data analysis validate the method's effectiveness for large-scale genomic experiments.

Conclusions:

• The robust meta-analysis method offers superior results for large-scale simultaneous testing in genomic experiments.
• It mitigates the issue of gross false discoveries often encountered with standard methods due to unobserved confounding variables.
• This approach is crucial for accurate interpretation of large genomic datasets.