Table 2.

Concordances with published calls (accuracies) or AWCA-based PAM50 calls for the main mLR classifiers.

Training set	Feature space of interest	Intended for	Accuracy on test set (%)	AWCA-PAM50 concordance on test set (%)	External test set	Accuracy on external test set (%)	AWCA-PAM50 concordance on external set (%)
${\mathit{TCGA}}_{\mathit{training}}$	PAM50	RSEM	89	89	PanCA	90	90
${\mathit{TCGA}}_{\mathit{training}}$	limma50	RSEM	88	87	PanCA	88	90
${\mathit{TCGA}}_{\mathit{training}}$	limma50_BWE	RSEM	87	87	PanCA	87	91
${GSE81538}_{\mathit{training}}$	PAM50	FPKM	93	92	GSE81538	92	93
${GSE81538}_{\mathit{training}}$	limma50	FPKM	91	91	GSE81538	89	89
${GSE81538}_{\mathit{training}}$	limma50_BWE	FPKM	91	91	GSE81538	89	89