Fig. 6

MA-plots of dilution data set before and after power-law correction. Fig. 6 shows the MA-plots (i.,e., average counts versus fold-changes) of the dilution dataset before (left-column) and after (right-column) the power-law correction. In particular, Figs a, b, c and d shows the MA-plot analysis for 4 mapping (Bowtie1, Bowtie2(global), Novoalign and BWA) algorithms while the permutation of the 6 normalization algorithms (DESeq, Relative Log Expression (RLE), Trimmed Mean of M-values (TMM), UpperQuartile (UQ), Count Per Million (CPM) and Quantile normalization) are arranged in a row-wise manner. For the power-law correction, the optimum PPS setting was evaluated to be 55 (See Additional file 6: Fig. S5A). In each MA-plot, the positive and noise signal are shown in red and blue respectively. The noise model (y = mx) is shown in dotted lines; Ideally, the slope value is 0 for no bias. The signal and noise residuals with respect to the noise model give the fold-change variation along the average count axis (or x-axis). Overall, it is apparent that the heteroskedasticity (see left-column) of the uncorrected AGS and NUGC3 count values has propagated down to the level of comparative analysis regardless of any combination of mapping and normalization methods. However when power-law correction is applied, heteroskedasticity was dramatically minimized (see right-column)