Table 2.

Performance of clinical response classifiers for clinically annotated gene expression datasets.

Dataset	Top 30 marker genes						Core marker genes
	SVM		FloWPS p = 0.95		FloWPS p = 0.90		SVM		FloWPS p = 0.95		FloWPS p = 0.90
	AUC	FDR	AUC	FDR	AUC	FDR	AUC	FDR	AUC	FDR	AUC	FDR
GSE25066 (Hatzis et al., 2011; Itoh et al., 2014)	0.70	0.28	0.76	0.10	0.77	0.13	0.73	0.26	0.76	0.25	0.76	0.23
GSE41998 (Horak et al., 2013)	0.79	0.25	0.87	0.14	0.91	0.14	0.87	0.14	0.89	0.15	0.92	0.12
GSE9782 (Mulligan et al., 2007)	0.73	0.28	0.78	0.22	0.76	0.17	0.68	0.33	0.71	0.33	0.72	0.34
GSE39754 (Chauhan et al., 2012)	0.65	0.36	0.68	0.27	0.71	0.34	0.65	0.36	0.68	0.36	0.72	0.35
GSE68871 (Terragna et al., 2016)	0.66	0.35	0.75	0.25	0.74	0.27	0.68	0.33	0.78	0.20	0.77	0.24
GSE55145 (Amin et al., 2014)	0.84	0.19	0.86	0.11	0.90	0.11	0.77	0.24	0.81	0.19	0.82	0.06
TARGET-50 (Goldman et al., 2015; Walz et al., 2015)	0.64	0.35	0.75	0.13	0.78	0.16	0.72	0.26	0.81	0.08	0.82	0.09
TARGET-10 (Goldman et al., 2015; Tricoli et al., 2016)	0.85	0.16	0.86	0.14	0.87	0.12	0.87	0.13	0.94	0.07	0.94	0.04
TARGET-20 (Goldman et al., 2015) with busulfan and cyclophosphamide	0.74	0.26	0.79	0.16	0.79	0.17	0.76	0.23	0.77	0.22	0.83	0.00
TARGET-20 (Goldman et al., 2015) w/o busulfan and cyclophosphamide	0.73	0.28	0.76	0.30	0.76	0.27	0.74	0.26	0.77	0.13	0.79	0.11

Area-under-curve (AUC) and false discovery rate (FDR) values calculated for each version of a classifier are shown. All calculations were made using leave-one-out cross-validation approach.