Good practice guidelines for biomarker discovery from array data: a case study for breast cancer prognosis

BMC Systems Biology

Table 3 Validation performance (AUROCC) of the candidate biomarkers in TRANSBIG data sets

Prognostic factors	TRANSBIG
	TDM at 5yrs		TDM at 10 yrs
	Node-	Node-/ER+	Node-	Node-/ER+
202705_at(CCNB2)	0.74	0.83	0.72	0.80
209642_at(BUB1)	0.71	0.81	0.70	0.78
204962_s_at(CENPA)	0.69	0.84	0.69	0.79
203362_s_at(MAD2L1)	0.68	0.75	0.67	0.71
202095_s_at(BIRC5)	0.67	0.78	0.65	0.74
210074_at(CTSL2)	0.65	0.64	0.65	0.64
209803_s_at(PHLDA2, TSSC3)	0.61	0.62	0.59	0.61
202338_at (TK1)	0.61	0.69	0.60	0.64
204086_at(PRAME)	0.61	0.62	0.57	0.58
202218_s_at (FADSD6)	0.50	0.49	0.50	0.45
210096_at(CYP4B1)	0.71	0.78	0.70	0.74
205883_at(ZNF145)	0.69	0.75	0.66	0.71
219197_s_at(SCUBE2, CEGP1)	0.66	0.66	0.63	0.59
214053_at(ERBB4)	0.66	0.72	0.67	0.74
208305_at(PGR)	0.65	0.66	0.64	0.66
219682_s_at(TBX3)	0.63	0.66	0.63	0.65
204541_at(SEC14L2)	0.63	0.65	0.62	0.59
206091_at(MATN3)	0.61	0.55	0.61	0.56
202554_s_at(GSTM3)	0.60	0.62	0.58	0.59
219440_at(RAI2)	0.59	0.59	0.56	0.56
Our 20-gene signature	0.73	0.83	0.70	0.79
16-gene signature	0.71	0.79	0.69	0.73
70-gene signature	0.68	NA	NA	NA
Nottingham Prognostic Index Score	0.67	0.68	0.66	0.66
Adjuvant! Online 10 year OS prob.	0.66	0.64	0.67	0.63
76-gene signature	0.65	0.68	0.62	0.64
Tumor grade	0.64	0.63	0.62	0.62
Tumor Size	0.63	0.65	0.63	0.64
212021_s_at(MKI67)	0.62	0.70	0.65	0.70
205225_at (ESR1)	0.58	0.59	0.57	0.61
Age	0.53	0.47	0.52	0.51

There are two AUROCC numbers for each gene at each endpoint. The first number is from the whole validation set with 100% Node- patients; the second is from the Node-/ER+ subset of the validation set. The numbers in bold font are significant at 95% confidence level. The top portion of the table contains 10 genes of one direction (over expression → poor prognosis). The middle portion contains 10 genes of the opposite direction (over expression → good prognosis). The bottom portion contains our signature based on all 20 genes and other prognostic factors. The performance of the 70-gene signature for the TRANSBIG data set is copied from [8]. The performance of the 76-gene signature is based on binary prediction of "good prognosis" and "poor prognosis" for each patient. Among the listed 20 genes, three genes (CENPA, GSTM3 and CEGP1) were included in the 70-gene signatures [9] and four genes (BIRC5, PGR, SCUBE2 and CTSL2) were included in the 16-gene signature. There is no overlap between our 20-gene signature and the 76-gene signa-ture.

ISSN: 1752-0509