Table 10.
ACT feature knock-out experiments for LR
| Features | F1 score | Specificity | Sensitivity | Accuracy | Matthews Coef | AUC iP/R |
|---|---|---|---|---|---|---|
| B | 72.33 | 91.61 | 68.28 | 84.84 | 62.14 | 78.97 |
| N | 50.05 | 94.10 | 38.20 | 77.88 | 40.75 | 60.12 |
| C | 69.38 | 89.39 | 66.91 | 82.87 | 57.58 | 76.30 |
| M | 69.61 | 90.83 | 65.37 | 83.44 | 58.53 | 75.06 |
| BC | 74.57 | 92.69 | 70.09 | 86.13 | 65.34 | 80.75 |
| BCM | 76.45 | 93.20 | 72.17 | 87.10 | 67.84 | 82.67 |
| BNCM | 76.78 | 93.49 | 72.23 | 87.33 | 68.37 | 82.89 |
Results of feature knock-out experiments on the combined ACT training and development datasets for the logistic regression (LR) model (%). B – bag of words; N – named entities; C – contextual words surrounding proteins; M – MeSH descriptors.