Can Zipf's law be adapted to normalize microarrays?

BMC Bioinformatics

Table 2 Kolmogorov-Smirnov values

	Microarray Platform	Pairwise Combinations (within array)	Raw	Global Median	Zipfs	Quantile	Loess	Loess HK	Zipfs Control	Zipfs HK
a.	Clontech Atlas Rat cDNA Expression	465	0.539	0.484	0.119	0	na	na	na	na
b.	Human Unigene RZPD 1	703	0.662	0.225	0.060	0	na	na	na	na
c.	Mouse Apoptosis	45 (5)	0.548 (0.631)	0.340 (0.318)	0.149 (0.167)	0 (0)	0.471 (0.042)	0.487 (0.172)	0.182 (0.179)	0.303 (0.296)
d.	Mouse Apoptosis Subset	28 (4)	0.568 (0.667)	0.303 (0.287)	0.111 (0.129)	0 (0)	0.317 (0.038)	0.341 (0.190)	0.145 (0.128)	0.315 (0.291)

Three microarray data sets presented in this paper and seven normalization techniques were compared by computing the mean Kolmogorov-Smirnov values of all possible pairwise combinations of arrays within a data set. In the case of the two channel mouse apoptosis microarray, within-array pairwise comparisons were also computed and are shown in parenthesis (here n = the number of arrays, as each array has 2 channels). The symbol 'na' indicates that the normalization techniques which can only be carried out on two channel (loess) or boutique (loess HK, Zipfs Control, Zipfs HK) arrays were not performed on single channel arrays. Values in bold typeface were computed for reference purposes only – these normalization methods cannot be validly applied to boutique microarrays.

ISSN: 1471-2105