Can Zipf's law be adapted to normalize microarrays?

BMC Bioinformatics

Table 1 Data set comparison

Set	Microarray Platform	Number of Data Points	Number of Expts	R2	GEO platform	GEO experiment	Array type
A.	Human Unigene RZPD 1	34560	31	0.9877	GLP284	GSE1510	cDNA, membrane
B.	Clontech Atlas Rat cDNA Expression	588	39	0.9968	GPL158	GSE1509	cDNA, membrane
C.	Clontech Atlas Human 1.2 (I & II)	1176	10	0.9903	GPL127, GPL128	GSE751	cDNA, membrane
D.	Clontech Atlas Mouse 1.2	1159	12	0.9460	GPL144	GSE565	cDNA, membrane
E.	Clontech Atlas Human Cancer 1.2	1160	36	0.9109	GPL158	GSE796	cDNA, membrane
F.	NlaIII: Rattus norvegicus	76790	1	0.9982	GPL23	GSM1679	SAGE
G.	NlaIII: Homo sapiens	101677	1	0.9978	GPL4	GSM14771	SAGE
H.	Mouse Apoptosis	1024	5 × 2	0.994	--	--	cDNA, glass
I.	Caltech 16K cDNA mouse	908	58	0.8892	na	na	cDNA, glass
J.	Stanford Human Unigene	908	24	0.9081	na	na	cDNA, glass
K.	Affymetrix GeneChip Rat Genome	8799	24	0.8538	GPL85	GSE776	Oligo, glass
L.	Affymetrix GeneChip Human Genome	12625	24	0.7773	GPL91	GSE803	Oligo, glass

Eleven microarray data set comparison. Raw intensities, without background subtraction, were used. Controls and blanks were excluded. For Affymetrix chips (K and L), MM/PM ratios were used. For data set B two different Atlas arrays were analyzed together, when analyzed separately they gave similar results. For two channel array systems (I and J), each channel was treated as a separate array. For set I, only the cyanine-3 channel (spleen sample control) was used and for set J, both channels were used for analysis. Reference for data set J: Ross et. al. [31].

ISSN: 1471-2105