Estimating the statistical significance of gene expression changes observed with oligonucleotide arrays

We present a simple method to assign approximate P-values to gene expression changes detected with Affymetrix oligonucleotide arrays and software. The method pools data for groups of genes and a small number of like-to-like comparisons in order to estimate the significance of changes observed for single genes in comparisons of experimental interest. Statistical significance levels are based on the observed variability in the fractional majority of probe pairs that indicate increasing or decreasing differential expression in comparisons of technical replicates. From this reference distribution or error model, we compute the expected frequency for fractional majorities in comparisons for N > or = 2. These computed distributions are the source of P-value estimates for changes seen in the experimental comparisons. The method is intended to complement the Affymetrix software and to rationalize gene selection for experimental designs involving limited replication.