Improved reproducibility achieved by means of predicting the noise-affected class of data by the parametric normalization. The same data sets shown in Fig. 7 were used. Each channel of data in a pair of dye-swapping hybridizations was normalized by means of the parametric method. Red and black dots indicate model-inconsistent and consistent data, respectively. In the control experiment, data ranking below -2.3 and above 1.8 were inconsistent with the distribution model, while those below -2.8 and above 1.9 in the treatment experiment were inconsistent with the distribution model. The 223 lowest-intensity data were excluded as "not detected" since they were found to be negative through γ subtraction. The rank vs. log-ratio plots present intensity dependence in terms of measured log-ratio (upper part of the panels; only one of the pair is shown). Scattergrams (lower part of the panels) present the reproducibility of the log-ratio measurements at the given range of intensity ranks. Data were obtained from dye-swap experiments ; the experiment IDs were 15972 and 15973 , and each channel was comprised of 11501 data. The rank vs. log-ratio plots presented above are from the ID 15973 data. The green and blue lines show the moving averages and moving standard deviations, respectively; the window of the calculation was 200.