Table 9.
ACT feature knock-out experiments for SVM
| Features | F1 score | Specificity | Sensitivity | Accuracy | Matthews Coef | AUC iP/R |
|---|---|---|---|---|---|---|
| B | 73.45 | 93.02 | 67.95 | 85.75 | 64.19 | 72.44 |
| N | 31.75 | 98.07 | 19.76 | 75.35 | 31.50 | 42.04 |
| C | 69.47 | 93.58 | 61.58 | 84.30 | 60.03 | 69.98 |
| M | 69.07 | 91.63 | 63.56 | 83.49 | 58.33 | 68.33 |
| BC | 74.93 | 94.10 | 68.55 | 86.69 | 66.50 | 73.92 |
| BCM | 76.71 | 94.33 | 70.86 | 87.52 | 68.70 | 76.00 |
| BNCM | 77.01 | 94.48 | 71.08 | 87.69 | 69.14 | 76.22 |
Results of feature knock-out experiments on the combined ACT training and development datasets (%) with Support Vector Machine (SVM). B – bag of words; N – named entities; C – contextual words surrounding proteins; M – MeSH descriptors.