Table 1 Overview of several pre-processing methods.

MAS

Global scaling – individual array normalization; moderate influence on expression levels, no effect on outliers. Non-parametric methods are potentially more reliable

Tukey biweight (robust average) – subtract MM from PM and adjust for negative values

dChip

Average median scaling – individual array normalization; moderate influence on expression levels, no effect on outliers. Non-parametric methods are potentially more reliable

Model-based index estimate – subtract MM from PM, but take indivual probe variability, assessed over all available arrays, into account

RMA

Quantile normalization – multiple array normalization; considerable influence on expression levels, with removal of outliers. Parametric methods are potentially more reliable

Median polish – only use MM for background adjustment; fit parameters of linear model robustly using median polish, taking into account all available arrays

GCRMA

Quantile normalization – multiple array normalization; considerable influence on expression levels, with removal of outliers. Parametric methods are potentially more reliable

Median polish – only use MM for background adjustment; fit parameters of linear model robustly using median polish, taking into account all available arrays; fit extra GC-content parameter