Normalization using ZScores

Sets of signal intensities are centered on zero with overall variance of one. The underlying assumption is that overall the logarithmic data follows a normal distribution which can be transformed into N(0,1) by log(X¡)

= (log x¡ - E(log x¡))/)(log x). Normally, as estimators for E, the mean, and for ), the standard deviation is used (for example, see (10)).

Scaling methods can only correct for systematic effects that are globally multiplicative. They were used in the very beginning of technology development (11, 12). Also, early analysis packages used scaling methods (AtlasImage 1.101, ClonTech).

0 0

Post a comment