Figure 1

Effect of different transcript concentrations illustrated by MvA plots. A) An MvA plots of mis-scaled data (top) and normal data (bottom). In the top plot, the combined effect of non-linear signal response and differences in transcript concentration causes the points to lie in curved region (replicate 2 is the darker microarray, both replicates taken 8 hours after irradiation). In contrast, when transcript concentrations are similar the cloud of points is not curved, but aligned with the horizontal axis (bottom plot, replicates 1 and 2 for the 0 hours time point). Focusing back to the top plot (region boxed in red), one can see how scaling mismatches occur in the median signal range: since the scaling performed in MAS5 takes into account almost all the signals, the signals of those genes lying in the median signal range are underscaled on the second chip: the median section of the cloud is underneath the horizontal axis. B) The scaling mismatch is even more striking when considering a simple time course application using the raw data. Within each replicate, intermediate time points have been linearly interpolated using neighbouring time points (e.g. for the 2 hours time point, the 0 and 4 hours chips have been used). The relative difference between those interpolation results and the actual values have been computed and the graphs show, for each intermediate time point, the first, second, and third quartile of the distribution of these relative differences. The scaling imbalance that occur with non-rescaled data is particularly visible for the 8 hours time point on the second replicate. To see how our algorithm corrects this problem, see Figure 6.