Unified Approach to False Discovery Rate Estimation
Author Information
Author(s): Korbinian Strimmer
Primary Institution: Institute for Medical Informatics, Statistics and Epidemiology, University of Leipzig
Hypothesis
Can a unified algorithm for estimating both local and tail area-based false discovery rates be developed?
Conclusion
The proposed procedure generalizes several specialized algorithms and offers a common framework for FDR estimation that performs comparably to the best existing methods.
Supporting Evidence
- The algorithm can be applied to various test statistics, including p-values and correlations.
- It allows for empirical null modeling, which accounts for dependencies among tests.
- The method is implemented in the R package 'fdrtool', available under the GNU GPL.
Takeaway
This study created a new method to help scientists figure out how many of their findings are real when they test many things at once.
Methodology
The study presents a semiparametric algorithm for estimating both local and tail area-based FDR using a modified Grenander density estimator.
Potential Biases
Potential bias if the null model is not correctly specified.
Limitations
The algorithm may not perform well if the distribution of observed p-values is misspecified.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website