Checking False Discovery Rates on Pvplots
Résumé
Under the null hypothesis, the p-value of a test is uniformly distributed. This is used for assessing the quality of a test, in particular its false detection rate. In order to enhance small p-values, which are of high practical importance, a plot in logarithmic coordinates, called pvplot, is proposed. The Goodness-of-Fit of the uniform distribution to a series of p-values is preferably tested using the Anderson-Darling test. The method is illustrated by simulated examples. A practical application to Gene Set Enrichment Analysis in genomic databases is described.