A robust tool for discriminative analysis and feature selection in paired samples impacts the identification of the genes essential for reprogramming lung tissue to adenocarcinoma

Table 1 Distribution of the number of false classifications of the 27 paired samples.

	False classifications	MWT	EDGE	PAM	Students t-test	Wilcoxon test
Up-regulated in tumours	0	1628	1109	938	1087	1054
	1	0	202	211	184	187
	2	0	27	73	20	14
	3	0	1	25	0	0
	4	0	0	6	0	0
	9	0	0	1	0	0
Down- regulated in tumours	8	0	0	1	0	0
	7	0	0	1	0	0
	6	0	0	2	0	0
	5	0	0	12	0	0
	4	0	3	29	2	2
	3	0	16	67	15	10
	2	0	64	118	68	64
	1	0	317	314	343	372
	0	1201	1090	1031	1110	1126
	# of the probe sets with 2 or more misclassified pairs	0	111	335	105	90
	total # of probe sets	2829	2829	2829	2829	2829
	percent of probe sets with 2 or more misclassified pairs	0	0.04	0.12	0.04	0.03

Comparison of the classification accuracy of highly discriminative 2829 probe sets (selected using MWT on cross-normalized signal intensity values with a stringent 100% accuracy criteria and a bootstrap p-value cut-off<0.05) with top-level 2829 probe sets identified by standard Wilcoxon sign ranked test (WT), EDGE, PAM, and student&#8217s t-test. While the ECD classifier was derived using cross-normalized dataset as input, the classifiers derived using PAM, EDGE, WT and t-test used the original MAS5-normalized data as input. However the classification accuracy in terms of the number of probe sets with 2 or more anomalous fold changes was estimated using the cross-normalized dataset.

ISSN: 1471-2164