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 2 The predictive performance of Multi-TGDR frameworks and PLS-DA

	# metabolites	Error on the data (%)	GBS	5-fold CV Error (%)
A. (without filtering)
Multi-TGDR global No Bagging	45	0	4.24e-05	3.82
Bagging (freq > 40%)	30	0	3.68e-05	3.82
Multi-TGDR local No Bagging	48	0	7.57e-05	5.34
Bagging (freq > 40%)	29	0	5.97e-04	6.11
B. (after moderated t-test filtering)
Multi-TGDR global No Bagging	42	0	1.03e-04	4.58
Bagging (freq > 25%)	37	0	1.13e-04	4.58
Bagging (freq > 40%)	26	0	3.58e-04	4.58
Multi-TGDR local No Bagging	42	0	6.18e-04	6.11
Bagging (freq > 25%)	38	0	6.87e-04	5.34
Bagging (freq > 40%)	25	0	2.24e-03	6.11
C. the performance of PLS-DA on the whole data
Naïve Bayes as the extra classifier	42	4.58	4.63e-02	7.63

A. The performance of multi-TGDR frameworks on the whole data: without moderated t-test filtering. B. The performance of multi-TGDR frameworks on the reduced data: with t-test filtering and 72 metabolites were filtered out. C. The performance of PLS-DA with naïve Bayes as the classifier. 42 metabolites selected by original analysis in Zhou’s study ref. [19] were used.
Note: For the reduced data, the optimal cutoff of bagging frequencies is 25%. However, in order to make a fair comparison with the results from the whole data, we analyzed the reduced data with bagging frequencies as 40% as well.

ISSN: 1471-2105