Table 2.
The performances of three de novo (SVM, KNN and PNN) and the deep learning (CNN) methods proposed in this study on the training and testing data sets (for eight representative GO families) with the highest similarity based on popular measurements (SE, SP, PR, AC and MCC). Each GO family was denoted by its GO ID, and its corresponding GO term was shown in Table 1
| GO ID | CNN | SVM | KNN | PNN | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SE% | SP% | PR% | AC% | MCC | SE% | SP% | PR% | AC% | MCC | SE% | SP% | PR% | AC% | MCC | SE% | SP% | PR% | AC% | MCC | |
| GO:0097472 | 85.2 | 99.9 | 99.1 | 98.5 | 0.91 | 81.0 | 99.1 | 90.2 | 97.3 | 0.84 | 70.9 | 97.7 | 76.4 | 95.1 | 0.71 | 72.7 | 98.6 | 84.7 | 96.2 | 0.76 |
| GO:0004721 | 79.5 | 99.2 | 93.9 | 96.6 | 0.85 | 65.2 | 99.3 | 93.2 | 94.7 | 0.75 | 68.0 | 95.6 | 70.8 | 91.9 | 0.65 | 55.5 | 98.7 | 87.0 | 92.9 | 0.66 |
| GO:0051347 | 81.3 | 99.6 | 97.5 | 96.9 | 0.87 | 78.4 | 97.3 | 83.4 | 94.5 | 0.78 | 75.6 | 93.5 | 66.9 | 90.9 | 0.66 | 63.8 | 99.1 | 92.8 | 93.9 | 0.74 |
| GO:0003735 | 80.7 | 98.7 | 95.8 | 93.8 | 0.84 | 87.2 | 97.6 | 93.2 | 94.8 | 0.87 | 84.7 | 95.6 | 87.8 | 92.6 | 0.81 | 84.9 | 95.5 | 87.7 | 92.6 | 0.81 |
| GO:0016746 | 71.7 | 96.2 | 85.8 | 90.2 | 0.72 | 70.5 | 96.6 | 86.8 | 90.2 | 0.72 | 70.7 | 92.5 | 75.2 | 87.2 | 0.65 | 64.8 | 93.7 | 76.8 | 86.7 | 0.62 |
| GO:0043085 | 69.3 | 98.1 | 94.3 | 89.2 | 0.74 | 75.4 | 95.2 | 87.5 | 89.1 | 0.74 | 78.8 | 87.1 | 73.2 | 84.5 | 0.65 | 66.7 | 95.6 | 87.2 | 86.7 | 0.68 |
| GO:0003700 | 81.1 | 95.0 | 92.3 | 89.0 | 0.78 | 82.8 | 94.8 | 92.2 | 89.7 | 0.79 | 81.9 | 90.9 | 87.0 | 87.1 | 0.74 | 84.3 | 89.0 | 85.0 | 87.0 | 0.73 |
| GO:0044093 | 80.7 | 93.5 | 86.6 | 89.1 | 0.76 | 78.7 | 94.1 | 87.5 | 88.8 | 0.75 | 80.6 | 85.5 | 74.5 | 83.8 | 0.65 | 72.6 | 92.5 | 83.7 | 85.6 | 0.68 |