Table 3.

Distribution of the null P-values for the simulated data before and after normalization.

Quantile	0.01	0.05	0.10	0.25	0.50
Raw	0	0	0	0	0
EigenMS	0	0.04	0.06	0.19	0.45
Global	0.34	0.69	0.83	0.96	1.00
Smoothing	0.35	0.68	0.78	0.89	0.99
ANOVA	0	0	0	0	0.03

Since none of the 120 peptides included in this comparison were differentially expressed, the distribution of P-values should be uniform between 0 and 1. Hence, for a given quantile, we expect that percentage of the 120 null peptides to be called significant. The raw data have null P-values that are extremely skewed left, due to strong unmodeled sources of bias. ANOVA normalization adjusts for batch effects and misses the strong additional technical features included in the simulation, resulting in a P-value distribution skewed like that in the raw data. Scatterplot smoothing and TIC normalization result in underestimated P-values and anticonservative inferences. EigenMS is approximately uniformly distributed as expected, being slightly conservative.