Figure 10

Variance as a function of intensity for Affymetrix preprocessing algorithms. The lowess trend curves of log₂$\widehat{\varphi }$ versus the average log-intensity of MCF7 and Jurkat for the three pre-processing algorithms MAS5 (red), PLIER (blue) and RMA (black). It can be seen that MAS 5 gives higher variance than the other algorithms over all intensities while RMA and PLIER are almost identical at high intensities. At low intensities the RMA algorithm is more precise than PLIER, perhaps at the cost of greater bias.