Table 3.
Performance comparison of various SVM models for predicting TAP binding affinity of peptides after jackknife testing
| Polynomial kernel | RBF kernel | |||
| SVM models | Parameters | Correlation coefficient (r) | Parameters | Correlation coefficient (r) |
| Only sequence based | C = 5.00 | 0.812 | C = 15.00 | 0.795 |
| D = 1 | G = 0.005 | |||
| Only features based (33 physicochemical) | C = 5.05 | 0.80 | C = 14.1 | 0.793 |
| D = 1 | G = 0.005 | |||
| Sequence + 33 features based (33) | C = 0.5 | 0.819 | C = 16.1 | 0.825 |
| D = 1 | G = 0.005 | |||
| Cascade SVM | ||||
| First modela (average results of 33 models) | C = 5.00 | 0.80 | ||
| D = 1 | ||||
| Second model | C = 1 | 0.86 | C = 30 | 0.88 |
| D = 3 | G = 2.0 | |||
The best value achieved using various approaches has been shown in bold.
a Average results of 33 models generated in first layer of SVM. Thirty-three models were generated by combining one feature of amino acids with sequence information each time.