Table 1.

Microarray data sets used for model development and validation in the MAQC-II project

Date set code	Endpoint code	Endpoint description	Microarray platform	Training seta				Validation seta				Comments and references
				Number of samples	Positives (P)	Negatives (N)	P/N ratio	Number of samples	Positives (P)	Negatives (N)	P/N ratio
Hamner	A	Lung tumorigen vs. non-tumorigen (mouse)	Affymetrix Mouse 430 2.0	70	26	44	0.59	88	28	60	0.47	The training set was first published in 2007 (ref. 50) and the validation set was generated for MAQC-II
Iconix	B	Non-genotoxic liver carcinogens vs. non-carcinogens (rat)	Amersham Uniset Rat 1 Bioarray	216	73	143	0.51	201	57	144	0.40	The data set was first published in 2007 (ref. 51). Raw microarray intensity data, instead of ratio data, were provided for MAQC-II data analysis
NIEHS	C	Liver toxicants vs. non-toxicants based on overall necrosis score (rat)	Affymetrix Rat 230 2.0	214	79	135	0.58	204	78	126	0.62	Exploratory visualization of the data set was reported in 2008 (ref. 53). However, the phenotype classification problem was formulated specifically for MAQC-II. A large amount of additional microarray and phenotype data were provided to MAQC-II for cross-platform and cross-tissue comparisons
Breast cancer (BR)	D	Pre-operative treatment response (pCR, pathologic complete response)	Affymetrix Human U133A	130	33	97	0.34	100	15	85	0.18	The training set was first published in 2006 (ref. 56) and the validation set was specifically generated for MAQC-II. In addition, two distinct endpoints (D and E) were analyzed in MAQC-II
	E	Estrogen receptor status (erpos)		130	80	50	1.6	100	61	39	1.56
Multiple myeloma (MM)	F	Overall survival milestone outcome (OS, 730-d cutoff)	Affymetrix Human U133Plus 2.0	340	51	289	0.18	214	27	187	0.14	The data set was first published in 2006 (ref. 57) and 2007 (ref. 58). However, patient survival data were updated and the raw microarray data (CEL files) were provided specifically for MAQC-II data analysis. In addition, endpoints H and I were designed and analyzed specifically in MAQC-II
	G	Event-free survival milestone outcome (EFS, 730-d cutoff)		340	84	256	0.33	214	34	180	0.19
	H	Clinical parameter S1 (CPS1). The actual class label is the sex of the patient. Used as a “positive” control endpoint		340	194	146	1.33	214	140	74	1.89
	I	Clinical parameter R1 (CPR1). The actual class label is randomly assigned. Used as a “negative” control endpoint		340	200	140	1.43	214	122	92	1.33
Neuro-blastoma (NB)	J	Overall survival milestone outcome (OS, 900-d cutoff)	Different versions of Agilent human microarrays	238	22	216	0.10	177	39	138	0.28	The training data set was first published in 2006 (ref. 63). The validation set (two-color Agilent platform) was generated specifically for MAQC-II. In addition, one-color Agilent platform data were also generated for most samples used in the training and validation sets specifically for MAQC-II to compare the prediction performance of two-color versus one-color platforms. Patient survival data were also updated. In addition, endpoints L and M were designed and analyzed specifically in MAQC-II
	K	Event-free survival milestone outcome (EFS, 900-d cutoff)		239	49	190	0.26	193	83	110	0.75
	L	Newly established parameter S (NEP_S). The actual class label is the sex of the patient. Used as a “positive” control endpoint		246	145	101	1.44	231	133	98	1.36
	M	Newly established parameter R (NEP_R). The actual class label is randomly assigned. Used as a “negative” control endpoint		246	145	101	1.44	253	143	110	1.30

The first three data sets (Hamner, Iconix and NIEHS) are from preclinical toxicogenomics studies, whereas the other three data sets are from clinical studies. Endpoints H and L are positive controls (sex of patient) and endpoints I and M are negative controls (randomly assigned class labels). The nature of H, I, L and M was unknown to MAQC-II participants except for the project leader until all calculations were completed.

^aNumbers shown are the actual number of samples used for model development or validation.