Table 5.
Prediction performance of logS and logD tasks based on embedding extracted from VAEs jointly trained with several descriptors
| Task | Descriptors | Classifier | ||
|---|---|---|---|---|
| 1D ResNet R2/RMSE |
MLP R2/RMSE |
LR R2/RMSE |
||
| logS | MolLogP | 0.790/0.926 | 0.770/0.971 | 0.751/1.010 |
| MolLogP, PEOE_VSA6 | 0.809/0.88 | 0.769/0.970 | 0.752/1.006 | |
| logP, PEOE_VSA6, MolWt | 0.804/0.896 | 0.771/0.967 | 0.762/0.986 | |
| logD | MolLogP | 0.520 / 0.840 | 0.319 / 1.001 | 0.296 / 1.018 |
| MolLogP, NumAromaticRings | 0.522 / 0.839 | 0.335/0.989 | 0.295/1.019 | |
|
MolLogP, NumAromaticRings, RingCount |
0.510/ 0.851 | 0.303/1.013 | 0.277/1.032 | |
Bold values indicate the best performance over all models