Orthogonal regression between qRT-PCR and normalization based log2 ratios. Regression of log2 ratios was conducted based on different normalization methods (y-axis) against qRT-PCR (x-axis). Equations and the respective regression lines are displayed in red. The grey dashed line indicates the main diagonal. Log2 ratios as calculated based on noBg_rankInvariant and noBg_log_loess pre-processing are overestimated in the lower and underestimated in higher ranges of log2 ratios. This over- and underestimation is more extreme for noBg_rankInvariant (intercept = 0.177, slope = 0.542) than for noBg_log_loess (intercept = 0.109, slope = 0.658). Data pre-processed using bg_rma_log_loess hardly over- or underestimates the data (intercept = -0.276, slope = 0.965). This results in scores of -2, 0, and 2 for noBg_rankInvariant, noBg_log_loess, and bg_rma_log_loess, respectively. An overview over the results for all pre-processing methods is given in Additional file 9.
Schmid et al. BMC Genomics 2010 11:349 doi:10.1186/1471-2164-11-349