Abstract:

Objective   To investigate the Bayesian interpretation of false discovery rate in multiple hypothesis testing and the empirical Bayes approach. Methods   A Bayesian two-group model was constructed for prostate cancer microarray data. The cumulative distribution function of the z value was empirically estimated and the density function was estimated using the method of Poisson regression by Efron, then the empirical Bayes approach was applied to estimate false discovery rate and local false discovery rate. Results   The false discovery rate was empirically estimated as 0-167 given the rejection region of (-∞,-3］, and the local false discovery rates were found to be no more than 0-2 in 58 genes. Conclusion   The false discovery rate can be interpreted from the Bayesian perspective and empirically estimated in high-dimensional data.

Key words:  Large scale data; Multiple hypothesis testing; False discovery rate; Bayesian and empirical Bayes approach; Density estimation; Prostate neoplasms