Figure 11

The distribution of expression measures obtained from intensity data shown in Fig. 1 and various preprocessing methods. The whole density distributions reveal subtle differences produced by the different methods. The distributions are computed separately for all probe sets (45,100) and for probe sets with at least two probes containing a (GGG)₁-motif (836, i.e. 2% of the total number) and for absent probe sets hybridized exclusively nonspecifically (45% of all probes). The multichip methods (RMA, gcRMA, dChip, vsn, Plier) are applied by computing intensity data of 5 arrays from the respective experimental series. 'log I' denotes the distributions of raw intensity data. The distributions of expression measures of probe sets containing (GGG)₁ probes for RMA, gcRMA and to a less degree for MAS5 and dChip are systematically shifted to the right compared with the distribution of all probe sets. These methods are partly unable to correct expression values for the (GGG)₁-bias whereas hook and Plier remove the bias.