Table 3 Performance comparison using sequence formulation techniques and hybrid feature vector
From: Deep-ProBind: binding protein prediction with transformer-based deep learning model
Methods | Dataset | ACC (%) | SN (%) | F1 (%) | SP (%) | MCC |
|---|---|---|---|---|---|---|
BERT | Training | 87.17 | 88.17 | 87.36 | 86.18 | 0.748 |
PsePSSM-DWT | 89.26 | 90.42 | 89.04 | 88.10 | 0.795 | |
Hybrid feature (without feature selection) | 90.28 | 91.14 | 90.36 | 89.35 | 0.812 | |
Hybrid features (with feature selection) | 92.67 | 93.41 | 93.40 | 91.82 | 0.853 | |
BERT | Independent | 91.20 | 91.91 | 90.91 | 90.50 | 0.824 |
PsePSSM-DWT | 91.49 | 92.47 | 91.46 | 90.47 | 0.830 | |
Hybrid feature (without feature selection) | 92.72 | 93.84 | 92.75 | 91.71 | 0.856 | |
Hybrid features (with feature selection) | 93.62 | 94.36 | 94.35 | 92.82 | 0.872 |