There are several well-known algorithms for preprocessing of microarray data, MAS.04, MAS.05, RMA, GCRMA, fRMA, and UPC.
Scientists believe that gcRMA's background adjustment introduces more noise than RMA into typical noisy chips produced in the lab .
We applied three popular normalization methods--RMA, GCRMA, and MAS5--with default setting.
In order to investigate the effect of normalization on the discriminative power of the pathway activity inference methods, we repeated this experiment using the USA and the Belgium datasets, where we first normalized the raw data using three different normalization methods (RMA, GCRMA, and MAS5) and then evaluated the discriminative power of the pathway markers.
For example, normalization of the Belgium dataset using GCRMA results makes the LLR method fail, as some of the genes loose variability and some of the LLR-values become infinite.
Next, we repeated the cross-dataset classification experiments based on the USA and the Belgium datasets after normalizing the raw data using RMA, GCRMA, and MAS5.