Welcome to the iEnhancer-DLRA code repository. iEnhancer-DLRA is a deep learning method based on a self-attentive mechanism to fuse local and global features of sequences to identify and classify enhancers.
The experimental results on the independent test dataset indicate that iEnhancer-DLRA performs better than nine existing state-of-the-art methods in both identification and classification of enhancers in all almost metrics.
Enhancer identifier :
| Method | ACC (%) | SP (%) | SN (%) | MCC |
|---|---|---|---|---|
| EnhancerPred | 74.00 | 74.50 | 73.50 | 0.480 |
| iEnhancer-2L | 73.00 | 75.00 | 71.00 | 0.460 |
| iEnhancer-EL | 74.75 | 78.50 | 71.00 | 0.496 |
| iEnhancer-5Step | 82.30 | 83.50 | 81.10 | 0.650 |
| iEnhancer-ECNN | 76.90 | 75.20 | 78.50 | 0.537 |
| iEnhancer-XG | 75.75 | 77.50 | 74.50 | 0.515 |
| iEnhancer-EBLSTM | 77.20 | 79.50 | 75.50 | 0.272 |
| iEnhancer-GAN | 78.40 | 75.80 | 81.10 | 0.567 |
| iEnhancer-RD | 78.80 | 76.50 | 81.00 | 0.576 |
| iEnhancer-DLRA | 93.72 | 90.45 | 97.00 | 0.876 |
| Improvement | +13.8% | +8.3% | +19.7% | +34.7% |
Enhancer classifier :
| Method | ACC (%) | SP (%) | SN (%) | MCC |
|---|---|---|---|---|
| EnhancerPred | 55.00 | 65.00 | 45.00 | 0.102 |
| iEnhancer-2L | 60.50 | 74.00 | 47.00 | 0.218 |
| iEnhancer-EL | 61.00 | 68.00 | 54.00 | 0.222 |
| iEnhancer-5Step | 63.50 | 74.00 | 53.00 | 0.280 |
| iEnhancer-ECNN | 67.80 | 56.40 | 79.10 | 0.368 |
| iEnhancer-XG | 63.50 | 57.00 | 70.00 | 0.272 |
| iEnhancer-EBLSTM | 65.80 | 53.60 | 81.20 | 0.324 |
| iEnhancer-GAN | 74.90 | 53.70 | 96.10 | 0.505 |
| iEnhancer-RD | 70.50 | 57.00 | 84.00 | 0.426 |
| iEnhancer-DLRA | 84.40 | 70.00 | 98.80 | 0.718 |
| Improvement | +12.6% | -5.4% | +2.8% | +42.1% |
- Python 3.7+
- Tensorflow 2.0+
git clone https://github.com/lftxd1/iEnhancer-DLRA.git
cd iEnhancer-DLRA
pip install -r requirements.txt
python benchmark_identifier.py
python benchmark_classifier.py
python rice_identifier.py