Table 3.
(a) Performance comparison by 5-fold cross-validation testing on the training data set of single-label proteins using SVMs; (b) Performance comparison of 5-fold cross-validation testing on the combined training data set (single- + dual-label) using SVMs; (c) Performance comparison of 5-fold cross-validation testing on the dual-localized training data set using SVMs. Bold values represents the best performance. RBF = radial basis function of SVM; C = regularization parameter.
| (a) | ||||
|---|---|---|---|---|
| Feature representation methods | Overall accuracy (%) (single-label data) | |||
| AAC (σ = 2, C = 10) | 73.65 | |||
| DIPEP (σ = 50, C = 500) | 77.56 | |||
| PseAAC (σ = 10, C = 500) | 75.49 | |||
| NCC (σ = 10, C = 50) | 74.36 | |||
| PseAAC-NCC-DIPEP (σ = 50, C = 500) | 81.97 | |||
| NCC-DIPEP (σ = 50, C = 500) | 81.18 | |||
| QSO (σ = 10, C = 500) | 73.25 | |||
| NCC-DIPEP-CTDC-CTDT- QSO (σ = 5, C = 30) | 80.42 | |||
| (b) | ||||
| Feature representation methods | Overall accuracy (%) (single- + dual-label data) | |||
| AAC (σ = 2, C = 10) | 68.48 | |||
| DIPEP (σ = 50, C = 500) | 74.59 | |||
| PseAAC (σ = 10, C = 500) | 71.87 | |||
| NCC (σ = 10, C = 50) | 70.74 | |||
| PseAAC-NCC-DIPEP (σ = 50, C = 500) | 84.75 | |||
| NCC-DIPEP (σ = 50, C = 500) | 83.96 | |||
| Physicochem [atomi + hydrophobicity, basic] | 73.21 | |||
| NCC-DIPEP-physicochem | 83.79 | |||
| Quasi-sequence-order descriptors | 54.38 | |||
| NCC-DIPEP-CTDC-CTDT- QSO | 60.02 | |||
| (c) | ||||
| Model | Kernel | C | Gamma | Overall accuracy (%) (dual- label data) |
| AAC | RBF | 10 | 0.001 | 76.64 |
| DIPEP | RBF | 10 | 0.001 | 82.29 |
| PseAAC | RBF | 10 | 0.001 | 77.63 |
| NCC | RBF | 10 | 0.001 | 86.02 |
| NCC-DIPEP | RBF | 10 | 0.001 | 87.57 |
| PseAAC-NCC-DIPEP | RBF | 10 | 0.001 | 87.88 |