Three-parameter lognormal distribution ubiquitously found in cDNA microarray data and its application to parametric data treatment

Table 2 Comparative table of normalization methods

	3-parameter lognormal method [this work]	LOWESS [2.3]	house-keeping genes	globalization [7]	A N O V A [9]	variance stabilization [5, 6]
assumed stable character	statistical data distribution	constant ratio tendency	expression levels of particular genes	sum of signal data	smallest differences in log(data)	smallest differences in arsinh(data)
Can the assumption be verified?	yes [22, 26] Figs. 1,2	no	yes¹	no	no	no
units for expression level	z-score	ratios to a reference	ratios to the stable genes	fraction (ppm)	ratios to a reference	statistical differences to a reference
data transformation for the adjustment	subtraction of a constant	non-linear	no	no	linear on logarithms	linear
numbers of data sets normalized in a calculation	1	2	1	1	all the sets to be compared	2
Can it compare multiple data sets without reference RNA?	yes	no	yes	yes	yes	no
amount of calculation	medium	medium	least	least	vast	large-medium
reproducibility	yes Figs. 3,8	no Fig. 7	nt³	no [7]	no²	no Fig. 6cd
Is the ratio tendency independent of signal intensity?	yes Figs. 5,8	yes Fig. 6b	nt³	no Fig. 6a	no²	yes Fig. 6cd
Is the ratio variance independent of signal intensity?	yes Figs. 5,8	no Fig. 6b[7]	nt³	no Fig. 6a	no²	yes Fig. 6cd
Can it find the level of additive noise in the data?	yes	no	no	no	no	no

¹A possible method is as follows: three or more gene candidates are chosen for the stable expression. Among the candidates, the ratio of every pair of data for a microarray experiment are calculated. If the expression levels are stable, the ratios are also stable. Such a combination of genes, however, has not been reported. ²Data are not presented. ³Since the stable genes could not be found among the data sets used in this article, this method was not tested (also see a review article [11]).

ISSN: 1471-2105