We tried to predict water solubility of molecules based on molecular structure. Graph models used are from machine learning library deepchem. We compared:
- Graph Convolutional Model
- Message Passing Neural Network
- Random Forest Regressor.
Dataset was split into training (80%) and testing (20%) set. Results were measured on testing set. Best performance was achieved by graph models especially MPNN.
Model | RMSE | MAE | R2 |
---|---|---|---|
GCM | 0.784 | 0.625 | 0.868 |
MPNN | 0.610 | 0.479 | 0.920 |
RFR | 1.142 | 0.872 | 0.701 |