Microarray Normalization Revisited for Reproducible Breast Cancer Biomarkers

Table 5

Performance of receptor status prediction after different normalization pipelines.

			Hormone receptor status prediction: percent misclassification against IHC
Samples		#	RmaMGlobal	GcrmaMGlobal	FrmaRSingle	IronGlobal	Mas5RSingle	PlierRSingle	RmaRSingle	RmaMSingle	GcrmaMSingle	GcrmaRGlobal	GcrmaRSingle	GcrmaRSingleCombat	Without normalization
ESR1	3014	2	8.0	8.0	8.3	8.9	9.0	11.5	10.3	8.5	10.1	7.9	9.3	19.9	12.0
PGR	2170	2	12.5	12.6	12.4	14.2	14.5	17.6	15.4	14.1	15.6	12.6	14.9	18.8	20.2
HER2	2443	2	8.5	8.5	8.8	9.8	9.0	11.1	9.7	9.6	9.7	8.8	9.9	15.4	10.9
Average			9.7	9.7	9.8	11.0	10.8	13.4	11.8	10.7	11.8	9.8	11.3	18.0	14.4

# denotes the number of levels for predictor (positive, negative); numbers are percentages of patients with predictions from gene expression different from the golden standard, IHC. The last column gives values for the use of raw, nonnormalized data from CEL-files. Results in terms of Matthews correlation coefficient are given in Table S3.