Figure 5

Effect of normalization methods on heteroscedasticity in the data. (A) The sorted mean intensity values for 1470 peaks are divided into six bins defined by the quantiles for the following cumulative probabilities: P= [0.025, 0.25, 0.5, 0.75, 0.975]. The first bin therefore contains the low abundance peaks on the left tail of the intensity distribution, with cumulative probability p < 0.025. The median coefficients of variance were calculated within each bin. (B) Scatter plot of unnormalized and NOMIS normalized peaks. The solid curves, shown for all four methods compared, are drawn as guides using robust locally weighted regression and smoothing scatter plots method (LOWESS) [22] with the 100 variable window size smoothing kernel.