Multi-TGDR, a multi-class regularization method, identifies the metabolic profiles of hepatocellular carcinoma and cirrhosis infected with hepatitis B or hepatitis C virus

Table 1 The comparison between Multi-TGDR frameworks and PLS-DA using simulated data

	# metabolites	Error on the data (%)	GBS	5-fold CV Error (%)
A. Simulation 1
Multi-TGDR: global No Bagging	105	0.76	0.0100	12.21
Global + Bagging (freq > 30%)	35	10.69	0.0773	12.21
local No Bagging	54 (14, 46)¹	2.29	0.0301	14.50
Local + Bagging (freq > 40%)	24 (14, 15)	8.40	0.0539	12.98
PLS-DA + Naïve Bayes as a classifier	89	14.50	0.1313	19.84
B. Simulation 2
Multi-TGDR: global No Bagging	110	0	0.0165	11.45
Global + Bagging (freq > 50%)	21	3.82	0.0237	7.63
local No Bagging	106(12, 95)	0	0.0067	9.16
Local + Bagging (freq > 40%)	25(9, 18)	3.82	0.0254	8.40
PLS-DA + Naïve Bayes as a classifier	97	6.87	0.1556	16.03

A. The performance of multi-TGDR frameworks and PLS-DA on the first simulated data. B. The performance of multi-TGDR and PLS-DA on the second simulated data.
¹(No.1, No.2): No.1 represents the number of metabolites selected in the first comparison (class 2 versus class 1) by multi-TGDR local. No.2 represents the number of metabolites selected in the second comparison (class 3 versus class 1).

ISSN: 1471-2105