The distribution of P-values in microarray experiments
by
Gary L. Gadbury
University of Missouri - Rolla

Microarray experiments commonly aim to identify differences in gene expression among groups of experimental units differing in genotype, age, diet, etc. In such experiments, statistical tests may be simultaneously conducted for each of several thousand genes. Recently some investigators have found that the distribution of p-values from these tests can contain useful information regarding several quantities of interest. These quantities are called true positive (TP), true negative (TN), and expected discovery rate (EDR). One method will be presented that models a continuous distribution of p-values and another that models a discrete distribution from randomization tests. Quantities of interest will be defined in terms of the former and the usefulness and limitations of both will be discussed. Simulations will assess the effect of certain patterns of correlation among gene expression levels on both the continuous and discrete models. A technique to evaluate the effects of varying sample sizes (i.e., number of arrays) on TP, TN, and EDR will be presented, time permitting.

Date received: August 11, 2003