Figure 22

Simulated dataset for method performance testing. Non-regular observations (k = 100) and regular observations (N-k = 6,068-100 = 5,968) as the main body of non-differentially-expressed genes were used for the simulations. Sample parameters are taken from the mac1 dataset. A random component was added to each outlier value using standard normal distribution with variance dependent on intensity. Heteroscedasticity for regular observations was also simulated by including intensity dependent variability in the low and high intensity levels. Three non-parametric smoothing methods were used to check absolute residuals for heteroscedasticity: supsmu, lowess and loess (the latter was based on a locally-quadratic fitting).