Table 4.
i6mA-CNN performance (marked in bold) comparison with state-of-the-art tools22.
| Dataset | Evaluation method | Tool | Sn | Sp | Acc | MCC | auROC |
|---|---|---|---|---|---|---|---|
| Dataset 1 | Jackknife testing | i6mA-Pred | 83.41% | 83.64% | 83.52% | 0.67 | 0.91 |
| i6mA-CNN | 90.61% | 94.12% | 93.15% | 0.85 | 0.98 | ||
| Dataset 1 | 10 Fold cross validation | iDNA6mA | 86.7% | 86.59% | 86.64% | 0.73 | 0.93 |
| i6mA-DNCP | 84.09% | 88.07% | 86.08% | 0.72 | 0.93 | ||
| MM-6mAPred | 89.32% | 90.11% | 89.72% | 0.79 | _ | ||
| 6mA-Finder | _ | _ | _ | _ | 0.94 | ||
| SDM6A | 85.2% | 90.9% | 88.1% | 0.76 | 0.94 | ||
| 6mA-RicePred | 84.89% | 89.66% | 87.27% | 0.75 | _ | ||
| DNA6mA-MINT | 94.25% | 90.8% | 92.53% | 0.85 | 0.95 | ||
| SpineNet-6mA (reproduced) | 86.48% | 92.39% | 89.43% | 0.79 | 0.95 | ||
| SpineNet-6mA (from reference) | 93.75% | 95.79% | 94.77% | 0.89 | 0.98 | ||
| i6mA-CNN | 90.35% | 94.62% | 92.48% | 0.85 | 0.98 | ||
| Dataset 2 | 5 Fold Cross Validation | iDNA6mA-Rice | 93% | 90.5% | 91.7% | 0.83 | 0.96 |
| SNNRice6mA | 94.33% | 89.75% | 92.04% | 0.84 | 0.97 | ||
| SpineNet-6mA (reproduced) | 93.94% | 91.36% | 92.15% | 0.86 | 0.97 | ||
| SpineNet-6mA (from reference) | 95.71% | 92.92% | 94.31% | 0.88 | 0.98 | ||
| i6mA-CNN | 95.13% | 92.81% | 93.97% | 0.88 | 0.98 |
Dataset 1 is the 880 sample per class dataset used for comparison purpose, while Dataset 2 is our benchmark dataset with 1,54,000 samples in each class. There are two results provided for SpineNet-6mA tool for both datasets—one obtained from reproducing the research and the other obtained from the reference research paper.