Empirical bayes methods and false discovery rates for microarrays

Area of Science:

• Bioinformatics
• Statistical genetics
• Genomics

Background:

• Microarray experiments generate large datasets with thousands of statistical tests.
• Simultaneous inference is a significant challenge in analyzing such high-dimensional data.
• Traditional two-sample tests, like Wilcoxon's statistic, are not directly applicable to massive gene sets.

Purpose of the Study:

• To explore and compare two distinct inferential approaches for simultaneous inference in microarray data analysis.
• To evaluate an empirical Bayes method and the frequentist false discovery rate approach.
• To demonstrate the relationship and potential combined use of these statistical methods.

Main Methods:

• An empirical Bayes method is presented, requiring minimal prior Bayesian modeling.
• The frequentist method of false discovery rates (FDR), as proposed by Benjamini and Hochberg (1995), is discussed.
• The study involves comparing the performance and characteristics of these two statistical techniques.

Main Results:

• The empirical Bayes method offers a practical approach to simultaneous inference with reduced prior assumptions.
• The false discovery rate method provides a robust framework for controlling errors in multiple testing scenarios.
• A close relationship between the empirical Bayes and false discovery rate methods is identified.

Conclusions:

• Both empirical Bayes and false discovery rate methods are effective for simultaneous inference in microarray studies.
• The two methods are closely related and can be integrated for improved statistical analysis.
• Combining these approaches allows for sensible simultaneous inferences in high-dimensional genomic data.