A direct approach to estimating false discovery rates conditional on covariates

Area of Science:

• Genomics
• Statistical genetics
• Bioinformatics

Background:

• Multiple hypothesis testing is prevalent in modern scientific research.
• Controlling error rates, particularly the false discovery rate (FDR), is crucial.
• Adaptive FDR methods require accurate estimation of the proportion of null hypotheses.

Purpose of the Study:

• To propose a novel regression framework for estimating the proportion of null hypotheses conditional on covariates.
• To develop a plug-in FDR estimator using this proportion and Benjamini-Hochberg adjusted p-values.
• To implement and evaluate the proposed method in a genome-wide association meta-analysis.

Main Methods:

• A regression framework is developed to estimate the proportion of null hypotheses.
• Covariates such as sample sizes and minor allele frequencies are utilized.
• The method is applied to a genome-wide association meta-analysis for body mass index.
• Simulation studies are conducted to evaluate the approach's performance.

Main Results:

• The proposed regression framework provides a conditional estimate of the proportion of null hypotheses.
• This estimate can be integrated with Benjamini-Hochberg adjusted p-values for improved FDR estimation.
• Application to a BMI genome-wide association meta-analysis demonstrates practical utility.
• Simulation results validate the effectiveness of the novel method.

Conclusions:

• The regression-based estimation of the null hypothesis proportion offers a flexible and powerful approach for FDR control.
• This method enhances the accuracy of FDR estimation by leveraging covariate information.
• The developed methodology is available as the swfdr package in Bioconductor for broader scientific application.