POWER_SAGE: comparing statistical tests for SAGE experiments

Motivation: The Serial Analysis of Gene Expression (SAGE) technology determines the expression level of a gene by measuring the frequency of a sequence tag derived from the corresponding mRNA transcript. Several statistical tests have been developed to detect significant differences in tag frequency between two samples. However, which one of these tests has the greatest power to detect real changes remains undetermined. Results: This paper compares three statistical tests for detecting significant changes of gene expression in SAGE experiments. The comparison makes use of Monte Carlo simulation that, in essence, generates 'virtual' SAGE experiments. Our analysis shows that the Chi-square test has the best power and robustness. Since the POWER-SAGE program can easily run 'virtual' SAGE studies with different combinations of sample size and tag frequency and determine the power for each combination, it can serve as a useful tool for planning SAGE experiments.