-log10(p-values) against MSQbetween where MSQbetween ≤ 5. MSQbetween was calculated based on the gene expression measured for the three sample groups analyzed, namely untreated HaCaT cells after 2, 4, and 12 hours. Results of three exemplary pre-processing methods of different quality are shown. bg_rma_log_loess exhibits the most unfavourable behaviour of the three. The p-values show a high variability over the whole range of MSQbetween and even for small MSQbetween values there are many relatively high -log10(p-values). Though for noBg_rankInvariant the p-values show less variability in general, especially for small MSQbetween values there are even more high -log10(p-values). In contrast, noBg_log_rsn exhibits less varying p-values and does not assign as many small p-values to low MSQbetween regions. The blue line represents a loess-curve fitted to the values. This curve takes uniformly larger values for noBg_rankInvariant and noBg_log_rsn than for bg_rma_log_loess indicating, on average, smaller p-values for the same MSQbetween value. Thus, in Figure 1, quality values of -1, 0, and 2 are assigned to bg_rma_log_loess, noBg_rankInvariant, and noBg_log_rsn, respectively. For an overview of all different normalization methods and their quality scores, see Additional file 1 and Figure 1.
Schmid et al. BMC Genomics 2010 11:349 doi:10.1186/1471-2164-11-349