Table 2.
Comparison of deep learning-based phosphorylation sites predictors.
| Tool name | Framework | Encoding strategy | Window size | Average AUC | Reference |
|---|---|---|---|---|---|
| MusiteDeep | Keras/TensorFlow | One-hot | 33 | 0.880 | [30] |
| PROSPECT | PyTorch | One-hot, EGAAC, CKSAAGP | 27 | 0.770 | [56] |
| DeepKinZero | TensorFlow | Word embedding | 15 | – | [60] |
| PhosTransfer | TensorFlow | Word embedding | – | 0.898 | [61] |
| GPS-PBS | Keras/TensorFlow | BLOSUM62 | 21 | 0.832 | [62] |
| DeepPPSite | Keras/TensorFlow | BE, EBGW, CKSAAP, PSPM, IPCP | 21 | 0.872 | [57] |
| DeepIPs | Keras/TensorFlow | Word embedding | 15 | 0.909 | [63] |
| PhosIDN | Keras/TensorFlow | One-hot, PPI embedding | 21 | 0.939 | [64] |
| EMBER | PyTorch | One-hot | 15 | 0.928 | [58] |
Note: -, data not available. AUC: Area under the Curve of ROC.