Using P-Values for the Planning and Analysis of Microarray Experiments
Microarray experiments to identify genes that change expression across multiple conditions can be used to gain clues about gene function. Many analysis strategies involve obtaining a p-value for a test of differential expression for each of thousands of genes. I will discuss an intuitively appealing method, originally proposed as an iterative algorithm by Mosig
et al. (2001, Genetics 157, 1683-1698), for estimating the total number of differentially expressed genes and the False Discovery Rate (FDR) associated with any threshold for statistical significance. I will characterize the limit of the iterative algorithm and describe how the estimator can be computed directly without iteration. I will compare the
performance of the resulting simple estimator with other procedures for estimating the number of true null hypotheses from a collection of observed p-values. I will conclude with a discussion of a new method that can provide information about power and sample size for a future experiment based on
p-values from a pilot experiment.