Related Experiment Videos
A Bayesian determination of threshold for identifying differentially expressed genes in microarray experiments.
1Merck Research Laboratories, P.O. Box 4, BL3-2, West Point, PA 19486, USA. jie_chen@merck.com
Statistics in Medicine
|December 14, 2005
Summary
This study introduces average false discovery rate (AFDR) and average false non-discovery rate (AFNR) by incorporating prior parameter distributions. These novel Bayesian methods offer more powerful and robust statistical inference than traditional FDR approaches.
Area of Science:
- Statistical Inference
- Bayesian Statistics
- Bioinformatics
Background:
- Traditional frequentist false discovery rate (FDR) and false non-discovery rate (FNR) are conditional on unknown parameters.
- Bayesian posterior FDR and FNR are conditional on observed data.
- A unified approach is needed to account for uncertainties in both parameters and data.
Purpose of the Study:
- To propose novel Bayesian metrics: average FDR (AFDR) and average FNR (AFNR).
- To introduce an overall risk measure, the average Bayes error rate (ABER).
- To develop formulas for AFDR, AFNR, and ABER for normal samples with hierarchical mixture priors.
Main Methods:
- Averaging frequentist risks (FDR, FNR) over prior parameter distributions.
- Developing formulas for AFDR, AFNR, and ABER.
- Applying methods to gene expression data for threshold selection.
- Conducting simulation studies to compare performance.
Main Results:
- Formulas for AFDR, AFNR, and ABER derived for specific prior distributions.
- Demonstrated threshold selection by minimizing ABER or controlling AFDR.
- Proposed methods showed improved power and robustness compared to standard FDR.
Conclusions:
- The proposed AFDR and ABER provide a more comprehensive risk assessment in statistical inference.
- These Bayesian-averaged methods enhance statistical power and robustness in hypothesis testing.
- The approach is particularly useful for analyzing complex biological data, such as gene expression.