Table 3. Ablation Evaluation of the Other MolHGT Componentsa.
data set | MolHGT | MolHGT-MR | MolHGT-EF | MolHGT + GRU |
---|---|---|---|---|
ESOL | 0.514 | 0.532 | 0.569 | 0.533 |
FreeSolv | 0.928 | 0.95 | 0.965 | 0.946 |
Lipo | 0.532 | 0.549 | 0.533 | 0.522 |
HIV | 0.779 | 0.789 | 0.787 | 0.778 |
BACE | 0.859 | 0.854 | 0.806 | 0.858 |
BBBP | 0.738 | 0.741 | 0.703 | 0.74 |
Tox21 | 0.862 | 0.863 | 0.866 | 0.865 |
ToxCast | 0.747 | 0.748 | 0.744 | 0.748 |
SIDER | 0.672 | 0.668 | 0.669 | 0.666 |
ClinTox | 0.901 | 0.91 | 0.898 | 0.909 |
The column “MolHGT” gave the results of the proposed model MolHGT on all the 10 data sets. The column “MolHGT-MR” gave the results of the MolHGT model without the metarelations. The column “MolHGT-EF” gave the data of the MolHGT model without the external features. Also, the column “MolHGT + GRU” gave the data of the MolHGT model with the original version of a gated recurrent unit (GRU) as the update function. The performance metrics RMSE and AUC were used for the first three regression data sets and the other seven classification data sets, respectively.