Estimating the score distributions of induced and non-induced genes in microarray experiments

A microarray experiment measures simultaneously the abundance of mRNA for several thousands of genes in a given tissue. With two classes of tissue or state of patient one can search for genes that show differential gene expression with respect to the classes. Based on the microarray data we assign a Wilcoxon ranksum score for differential gene expression to every gene. Typically a high score corresponds to overexpression and a low score to underexpression. A gene showing differential gene expression is called “induced”. The score distributions of microarray experiments are mixtures of score distributions of induced and non-induced genes. The two distributions can be distinguished if the truly induced genes show very high (or low resp.) scores. However, in a typical setting the two distributions overlap and only highly induced genes can be selected with common methods. The main part of the induced genes merge with non-induced genes in “twilight zones”. It is a challenging problem to discover where twilight zones are and how many truly induced genes one can expect at a certain score level. If we observe a score for a specific gene we cannot only tell how deep it is situated in a twilight zone but also how likely it is to be an induced gene if the mixture is known at that score level.