Identification of a minimum number of genes to predict triple-negative breast cancer subgroups from gene expression profiles

Table 3 Per-subgroup prediction ROC scores for up- and downregulated genes, before and after attribute selection

	BL1	BL2	M	IM	MSL	LAR	Weighted average	Validation option
	Per-subgroup prediction ROC metric before attribute selection (Total number of upregulated genes = 120)
Upregulated genes	0.979	0.97	0.983	0.992	0.996	0.999	0.987	Cross-validation on GEO set (tenold)
	0.862	0.949	0.837	0.890	0.904	0.784	0.852	Validation set: Italian set
	0.958	0.981	0.958	0.958	0.816	0.92	0.948	Validation set: TCGA set
	Per-subgroup prediction ROC metric after attribute selection (Total number of upregulated genes = 103)
	0.975	0.978	0.984	0.988	0.996	1	0.986	Cross-validation on GEO set (tenfold)
	0.916	0.955	0.843	0.922	0.962	0.7	0.883	Validation set: Italian set
	0.96	0.955	0.974	0.951	0.801	0.864	0.94	Validation set: TCGA set
	Per-subgroup prediction ROC metric before attribute selection (Total number of downregulated genes = 81)
Downregulated genes	0.986	0.977	0.987	0.991	0.988	0.998	0.988	Cross-validation on GEO set (tenfold)
	0.727	0.808	0.924	0.871	0.897	0.886	0.858	Validation set: Italian set
	0.678	0.788	0.861	0.781	0.858	0.985	0.813	Validation set: TCGA set
	Per-subgroup prediction ROC metric after attribute selection (Total number of downregulated genes = 77)
	0.984	0.961	0.984	0.986	0.985	0.996	0.984	Cross-validation on GEO set (tenfold)
	0.742	0.962	0.816	0.815	0.776	0.91	0.83	Validation set: Italian set
	0.743	0.807	0.813	0.776	0.7	0.977	0.802	Validation set: TCGA set

ISSN: 1479-7364