Smaller experiments benefit from filtering with a threshold of 0.5 or higher (a higher threshold of fraction Present); criteria can be relaxed for larger experiments although filtering is still of substantial value. Setting the threshold to 100% Present (all probe sets Present in at least one treatment group) is too stringent for experiments of any size, because it removes many of the genes being turned on or off and removes a large proportion of very significant genes with little improvement in FDR.
