The spike-in control based normalization (SCN) method. (A) Plot of the raw intensities for the 10 spike-in control probe sets “a”… “j” from the 16 arrays of the Exiqon lung dataset. The probe set intensity values were computed as the median values of the corresponding 48 probe intensities and the error bars correspond to the 2.5th–97.5th percentiles of the corresponding distributions. For a given spike-in control probe set, the “coherent deviations” mentioned in the text are estimated by the size of the between-array range of intensity values, divided by the between-array mean intensity value. (B) Heat map of the Pearson correlation matrix between all pairs of spike-in control probe sets shown in panel A. (C) Ratio between the variance of the raw and normalized intensities for the 10 spike-in control probe sets shown in panels A and B. (D) Raw intensity dependence of the spike-in controls based normalization intensity correction function ΔE computed for the 16 arrays of the lung dataset. The continuous curves represent the intensity correction function ΔE(x,k) defined for the continuous raw intensity values x given by the horizontal axis and the 16 discrete array labels k depicted in the color legend, while the points correspond to normalization intensity corrections ΔE(S)
for the 10 spike-in control probe sets j and the 16 array labels k (see the “Spike-in controls based normalization method” section in “Methods”).