Table 4.

Classification results of the RRHGE gene signature and other existing gene signatures on two testing sets, for example, (A) the Desmedt dataset and (B) the van de Vijver dataset.

	Algorithm	N	TP	FN	TN	FP	SN	SP	ACC	MCC
(A) Desmedt	GGI	190	84	45	29	32	0.651	0.475	0.595	0.121
	70 g	190	53	76	27	34	0.411	0.443	0.421	−0.137
	76 g	190	78	51	23	38	0.605	0.377	0.532	−0.018
	ITI	190	95	34	33	28	0.736	0.541	0.674	0.271
	HRGE	190	115	14	36	25	0.891	0.590	0.795	0.511
	RRHGE-H	190	103	26	46	15	0.798	0.754	0.784	0.532
	RRHGE-HI	190	100	29	48	13	0.775	0.787	0.779	0.535
	RRHGE-TSN	190	119	10	54	7	0.922	0.885	0.911	0.798
	RRHGE	190	123	6	56	5	0.953	0.918	0.942	0.868
(B) van de Vijver	GGI	150	77	37	17	19	0.675	0.472	0.627	0.131
	70 g	150	71	43	19	17	0.623	0.528	0.600	0.131
	76 g	150	72	42	20	16	0.632	0.556	0.613	0.162
	ITI	150	59	55	19	17	0.518	0.528	0.520	0.039
	HRGE	150	70	44	20	16	0.614	0.556	0.600	0.146
	RRHGE-H	146	92	22	14	18	0.807	0.438	0.726	0.235
	RRHGE-HI	150	94	20	22	14	0.825	0.611	0.773	0.414
	RRHGE-TSN	150	101	13	26	10	0.886	0.722	0.847	0.592
	RRHGE	150	105	9	28	8	0.921	0.778	0.887	0.692

Here, N defines the total number of samples, TP defines true positive (ER+ samples predicted as ER+), TN defines true negative (ER− samples predicted as ER−), FP defines false positive (ER− samples predicted as ER+), FN defines false negative (ER+ samples predicted as ER−), SE defines sensitivity, SP defines specificity, ACC defines accuracy, and MCC defines Matthews coefficient correlation. For simplicity, we represent the Genomic Grade Index as GGI, 70 gene signature as 70 g, 76 gene signature as 76 g, Interactome-Transcriptome Integration as ITI, and Hub-based Reliable Gene Expression as HRGE.. The RRHGE subnetwork based gene signature provides superior performance in both (A) Desmedt and (B) van de Vijver dataset.