The effect of correlation in false discovery rate estimation.

Area of Science:

• Statistical analysis
• Genomics
• Bioinformatics

Background:

• False discovery rate (FDR) is crucial for multiple hypothesis testing.
• Standard FDR analysis often assumes independence, which may not hold in real-world data.
• Understanding the impact of correlation is vital for accurate statistical inference.

Purpose of the Study:

• To quantify the effect of correlation on the false discovery rate estimator.
• To derive approximations for the mean, variance, distribution, and quantiles of the FDR estimator under correlation.
• To investigate the consistency of the FDR estimator with correlated data.

Main Methods:

• Utilized a negative binomial model for false discoveries.
• Estimated model parameters empirically from the data.
• Derived analytical approximations for FDR estimator properties.

Main Results:

• Correlation can substantially increase the bias and variance of the FDR estimator compared to independent data.
• The FDR estimator's consistency can fail under certain correlation structures, like exchangeable correlation.
• Approximations for mean, variance, distribution, and quantiles were derived for correlated data.

Conclusions:

• Correlation is a critical factor that must be accounted for in FDR analysis.
• Ignoring correlation can lead to biased and unreliable FDR estimates.
• The findings highlight the limitations of standard FDR methods with dependent test results.