Background: When analysing data which simultaneously implicate a large number of statistical tests, one of the main problems is to take into account the multiplicity of these tests. The huge amount of comparisons done in studies of which the aim is to detect, using microarray, genes whose transcriptional changes are related to a biological or clinical outcome leads to a renewed interest for the multiple comparisons problem. However, this problem concerns many other fields such as psychometry, epidemiology, genetics.
Results: First, we introduce the multiple comparison framework. Then, we present the main procedures based on a global error rate called the "Family Wise Error Rate" (FWER) and those based on a false discovery rate expectancy (the "False Discovery Rate" (FDR) and the "positive False Discovery Rate" (pFDR)). Next, we apply the different procedures on a real dataset from a breast carcinoma study. Finally, we discuss the main results and we present guidelines for the use of these procedures.