Selecting a single model or combining multiple models for microarray-based classifier development? – A comparative analysis based on large and diverse datasets generated from the MAQC-II project

Table 1 The datasets used in MAQC-II project.

Endpoint code	Endpoint	Endpoint description	Training set		Validation set
			#Sample	P/N ratio*	#Sample	P/N ratio
A	Lung tumorigenicity	Lung tumorigen vs. non-tumorigen	70	0.59	88	0.47
B	Non-genotoxicity	Non-genotoxic hepatocarcinogen vs. non-carcinogen	216	0.51	201	0.4
C	Liver toxicity	Liver toxicants vs. non-toxicants	214	0.58	204	0.62
D	Breast cancer	Pathologic complete response, pCR	130	0.34	100	0.18
E	Breast cancer	Estrogen receptor status (ER +/-)	130	1.6	100	1.56
F	Multiple myeloma	Overall survival	340	0.18	214	0.14
G	Multiple myeloma	Event-free survival	340	0.33	214	0.19
H	Multiple myeloma	Male vs. female (positive control)	340	1.33	214	1.89
I	Multiple myeloma	Random 2-class label (negative control)	340	1.43	214	1.33
J	Neuroblastoma	Overall survival	238	0.1	177	0.28
K	Neuroblastoma	Event-free survival	239	0.26	193	0.75
L	Neuroblastoma	Male vs. female (positive control)	246	1.44	231	1.36
M	Neuroblastoma	Random 2-class label (negative control)	246	1.44	253	1.36

* P/N = Positive/Negative ratio. Positive denotes for these samples showing the positive results (e.g. cancer, tumor).

ISSN: 1471-2105