Swirl dataset

Since the dye-swap dataset shows no nonlinearity, we chose a second dataset with a typical nonlinearity in the logarithmic scale. We used the first experiment of the swirl dataset that consists of one pair of RNA samples (wildtype and swirl). Background-subtracted intensity values were normalized by applying global mean scaling, global linear regression, loess regression, quantile normalization and qspline normalization.

For this dataset, local regression should out-perform global linear regression methods. The result is shown in Figure 17.3. Already from the visual inspection it is obvious that the application of local regression is more appropriate than applying a global normalization method. Quantile normalization and qspline normalization also seem to be superior to global methods in terms of correction of nonlinearities.

0 0

Post a comment