Table 1.

Characteristics of the breast cancer datasets used in this study

Dataset	Training/validation	Platform	Number of samples	ER+	Her2+	BasalL	Her2 E	LumA	LumB	NormaL
expO Bittner M. (www.intgen.org)	Training	Affymetrix (U133 Plus 2.0)	312	65.7%	28.1%	21.2%	16.3%	31.4%	18.9%	12.2%
Lu et al. Breast Cancer Res Treat 2008 [40]	Training	Affymetrix (U133 Plus 2.0)	127	58.3%	23.6%	26.8%	17.3%	37.0%	16.5%	2.4%
Li et al. Nat Med 2010 [41]	Training	Affymetrix (U133 Plus 2.0)	115	60.9%	31.3%	27.0%	16.5%	36.5%	18.3%	1.7%
Parker et al. J Clin Oncol 2009 [25]	Training	Agilent	226	58.2%	12.4%	31.0%	12.4%	33.2%	16.4%	7.1%
Curtis et al. Nature 2012 (METABRIC) [13]	Training	Illumina (HT-12 v3)	1992	76.2%	12.5%	20.5%	16.0%	26.7%	22.8%	14.0%
Guedj et al. Oncogene 2012 [42]	Training	Affymetrix (U133 Plus 2.0)	537	75.9%	13.0%	16.2%	17.1%	24.8%	24.2%	17.7%
TCGA Nature 2012 [23]	Training	Agilent	233	79.3%	21.9%	22.3%	15.5%	30.9%	21.0%	10.3%
Miller et al. PNAS 2005 [43]	Training	Affymetrix (U133AB)	251	86.2%	13.1%	15.9%	18.3%	25.1%	20.3%	20.3%
Pawitan et al. Breast Cancer Res 2005 [44]	Training	Affymetrix (U133AB)	159	N/A	13.8%	12.6%	13.8%	28.3%	27.7%	17.6%
TCGA Nature 2012 [23]	Training	RNA-seq (Illumina)	558	77.9%	24.2%	19.2%	12.9%	30.5%	22.2%	15.2%
McGill [12]	Validation	Affymetrix Gene ST	429	78.1%	18.5%	20.5%	17.4%	37.6%	16.2%	8.1%
TCGA Nature 2012 [23]	Validation	RNA-seq (Illumina)	313	72.6%	11.9%	19.2%	13.4%	35.8%	15.3%	16.3%

ER+ estrogen-receptor-positive, Her2+ human epidermal growth factor receptor 2-positive, BasalL basal-like intrinsic subtype, Her2E Her2-enriched intrinsic subtype, LumA luminal A intrinsic subtype, LumB luminal B intrinsic subtype, NormaL normal-like intrinsic subtype, RNA-seq RNA sequencing, METABRIC Molecular Taxonomy of Breast Cancer International Consortium, TCGA The Cancer Genome Atlas