Model selection and efficiency testing for normalization of cDNA microarray data

Table 3 Statistical comparison of normalization schemes

	Experiment	No normalization	Linear normalization	Global lowess	P-lowess	S-lowess	LIN	OLIN	OSLIN
Intensity-dependent cor(M, )	SW480/620	0.50	0.50	0.01	-0.01	-0.01	0.04	0.00	0.00
Intensity-dependent cor(M, )	Apo AI	0.47	0.47	0.06	0.05	0.05	0.09	0.00	0.00
Spatial cor(M, )	SW480/620	0.53	0.53	0.56	0.34	0.32	0.27	0.07	0.08
Spatial cor(M, )	Apo AI	0.58	0.58	0.59	0.41	0.38	0.43	0.15	0.15
Mean pairwise cor(M)	SW480/620	0.46	0.46	0.50	0.59	0.59	0.63	0.67	0.64
Var(M)	SW480/620	927.1	658.7	455.4	216.3	213.0	205.7	163.0	186.5
Number of significant genes	SW480/620	26	71	51	75	88	94	99	129
Average cor(M_qPCR, M_ma)	Fibroblast	0.82	0.81	0.82	0.82	0.81	0.81	0.84	0.88

Intensity-dependent correlation (intensity-dependent cor(M, )) describes the correlation between the log ratio M of the spot and the median value of M within a symmetrical neighborhood of 50 spots on the intensity scale. Spatial correlation (spatial cor(M, )) describes the correlation between the log ratio M of the spot and the median value of M within a neighborhood defined by a window of 5 × 5 spots. To ensure independence, M of the spot was not included in the median of the neighborhood. For the calculation of mean pairwise correlation of slides, spots with intensity A < 11.6 were excluded. The significance of differential gene expression was examined by a one-sample t-test with the null hypothesis of mean log ratio M = 0. Duplicated spots on SW480/620 arrays were treated as independent measurements producing a maximum of eight observations per gene. Genes were detected as significantly differentially expressed if their Bonferroni adjusted p-values were smaller than 0.01. For the fibroblast experiment, the average Pearson correlation between qPCR-based logged fold changes M_qPCRand microarray-based logged fold changes M_maof the genes for IL-8, COX2, Mast, B4-2 and actin is shown.

ISSN: 1474-760X