Table 1.
Assessment of model performance during model training and testing. During training, we monitored model performance with respect to different sampling strategies used during dataset assembly. With “extended uniform sampling” yielding the best results, we built DeepSTABp and evaluated its performance in direct comparison to ProTstab2.
| Metrics | Training (DeepSTABp) | Testing | |||
|---|---|---|---|---|---|
| Naïve | Uniform Sampling | Extended Uniform Sampling | DeepSTABp | ProTstab2 | |
| R2 | 0.86 | 0.89 | 0.93 | 0.80 | 0.57 |
| PCC | 0.93 | 0.96 | 0.97 | 0.90 | 0.76 |
| MAE (°C) | 3.20 | 2.43 | 1.81 | 3.22 | 4.95 |
| MSE (°C) | 17.84 | 9.70 | 5.54 | 18.46 | 41.59 |
| RMSE (°C) | 4.22 | 3.11 | 2.35 | 4.30 | 6.45 |