Table 2.
Comparison of CNN Classifiers of Different Feature Sets By 10 Times 10-Fold Cross-Validation
| Feature Set {#} | ACC | AUC-ROC | AUC-PR | Kappa | Sn | Sp | MCC |
|---|---|---|---|---|---|---|---|
| T {21} | 71.22 ± 0.51 | 77.41 ± 0.22 | 73.97 ± 0.53 | 42.43 ± 1.01 | 78.22 ± 1.1 | 64.21 ± 0.91 | 42.86 ± 1.06 |
| C {21} | 72.65 ± 0.35 | 78.33 ± 0.12 | 75.54 ± 0.32 | 45.30 ± 0.69 | 77.85 ± 1.66 | 67.45 ± 1.29 | 45.57 ± 0.78 |
| CTD {147} | 73.71 ± 0.34 | 79.96 ± 0.21 | 76.61 ± 0.48 | 47.41 ± 0.67 | 79.05 ± 1.93 | 68.36 ± 1.51 | 47.71 ± 0.79 |
| D {105} | 73.74 ± 0.23 | 79.92 ± 0.17 | 76.73 ± 0.30 | 47.48 ± 0.46 | 79.33 ± 1.2 | 68.14 ± 1.01 | 47.79 ± 0.52 |
| AAC {20} | 74.27 ± 0.26 | 80.48 ± 0.19 | 77.52 ± 0.31 | 48.55 ± 0.51 | 80.92 ± 0.85 | 67.63 ± 1.05 | 48.99 ± 0.48 |
| SC-PseAAC {32}a | 75.62 ± 0.27 | 82.07 ± 0.19 | 79.04 ± 0.37 | 51.24 ± 0.54 | 82.33 ± 0.78 | 68.91 ± 1.18 | 51.72 ± 0.45 |
| Five-best PseKRAAC {86} | 76.50 ± 0.37 | 82.48 ± 0.20 | 79.55 ± 0.5 | 53.00 ± 0.74 | 83.35 ± 0.86 | 69.65 ± 0.67 | 53.51 ± 0.78 |
Values shown are mean ± SD (values were multiplied by 100).
a
Parameters used for SC-PseAAC (series-correlation pseudo amino acid composition, commonly known as type-2 PseAAC) are λ = 4 and w = 0.2.