Table 4.

(a) Comparison of prediction results on an ‘independent data set’ based on models trained from single-label proteins using SVMs; (b) Comparison of prediction results on an ‘independent data set’ based on models trained from combined data set (single- + dual-label); (c) Comparison of prediction results on an ‘independent data set’ based on models trained from dual-label proteins data set. Bold values represents the best performance.

(a)
Feature representation methods	Accuracy (%)
AAC (σ = 2, C = 10)	59.11
DIPEP (σ = 50, C = 500)	59.11
PseAAC (σ = 10, C = 500)	59.12
NCC (σ = 10, C = 50)	50.34
PseAAC-NCC-DIPEP (σ = 50, C = 500)	64.36
NCC-DIPEP (σ = 50, C = 500)	64.05
QSO (σ = 10, C = 500)	57.05
NCC-DIPEP-CTDC-CTDT-QSO (σ = 5, C = 300)	61.46
(b)
Feature representation methods	Accuracy (%)
AAC (σ = 2, C = 10)	57.71
DIPEP (σ = 50, C = 500)	58.95
PseAAC (σ = 10, C = 500)	56.60
NCC (σ = 10, C = 50)	52.88
PseAAC-NCC-DIPEP (σ = 50, C = 500)	64.84
NCC-DIPEP (σ = 50, C = 500)	64.42
Quasi-sequence-order descriptors	58.94
NCC-DIPEP-CTDC-CTDT-QSO	38.49
(c)
Model	Kernel	C	Gamma	Accuracy (%)
AAC	RBF	10	0.001	72.56
DIPEP	RBF	10	0.001	72.97
PseAAC	RBF	10	0.001	75.67
NCC	RBF	10	0.001	78.37
NCC-DIPEP	RBF	10	0.001	75.67
PseAAC-NCC-DIPEP	RBF	10	0.001	81.08