Many accurate small-discriminatory feature subsets exist in microarray transcript data: biomarker discovery

BMC Bioinformatics

Table 2 Number of 1-, 2- and 3-LCs for the data sets described in Table 1. For the two-class partitionings shown, the Table gives the number of perfect linear classifiers (given P initial features) that can be constructed using one, two and for some sets 3, genes. The 3 gene results report the total number of triples, and in parenthesis the number saved for later analysis.

Data set	Class +1	Class -1	1-LCs	2-LCs	3-LC
BreastER	ER⁺	ER^-	0	16	108 k(2045/180¹)
BreastBRCA	BRCA1	BRCA2	18	143,574
	BRCA1/BRCA2	Sporadic	0	2,114
	BRCA1	BRCA2/Sporadic	0	12,729
	BRCA1	Sporadic	4	66,754
	BRCA2	BRCA1/Sporadic	0	10,027
	BRCA2	Sporadic	7	78,901
OvarianBRCA	BRCA1	BRCA2	0	1,612
	BRCA1/BRCA2	Sporadic	0	0
	BRCA1	BRCA2/Sporadic	0	0
	BRCA1	Sporadic	0	0	249²
	BRCA2	BRCA1/Sporadic	0	0
	BRCA2	Sporadic	0	0	2³
LungStanford	AC	SCC	2	484	565 k(65)
LungBeer	AC	Normal	5	22,102
Cutaneous	Melanoma	Cell line	0	596	4.2 m(386)
GIST	KIT ⁺	Spindle	74	137,981
YeohALL	T	E2A/BCR/TEL/MLL/Hdip50	1	1169
	E2A	T/BCR/TEL/MLL/Hdip50	4	386
	BCR	T/E2A/TEL/MLL/Hdip50	0	1
	TEL	T/E2A/BCR/MLL/Hdip50	0	3
	MLL	T/E2A/BCR/TEL/Hdip50	0	2
	Hdip50	T/E2A/BCR/TEL/MLL	0	0
Prostate	Tumor	Normal	52	249,665
Liver	Tumor	Normal, Original Labels	0	0	44⁴
	Tumor	Normal, Relabeled	0	4	9.5 k(291⁴)

(¹Only 180 of the estrogen triples do not contain the Estrogen Receptor 1 gene. ²Ovarian BRCA B1 vs Sporadic triples were run with a reduced set of genes, variation filtered to 2109. ³Ovarian BRCA B2 vs Sporadic triples were run with a reduced set of genes, variation filtered to 2097. ⁴Liver Cancer triples use only 1956 genes. Relabeled means the the 2 "outlier" samples are re-labeled as normal.)

ISSN: 1471-2105