Table 2 Overall performance comparison to the state-of-the-art methods on molecular property prediction regression tasks.
From: Pharmacophoric-constrained heterogeneous graph transformer model for molecular property prediction
Regression (RMSE, lower is better↓) | ||||||
|---|---|---|---|---|---|---|
Dataset | ESOL | FreeSolv | Lipophilicity | ESOL | FreeSolv | Lipophilicity |
Molecules | 1128 | 642 | 4200 | 1128 | 642 | 4200 |
Tasks | 1 | 1 | 1 | 1 | 1 | 1 |
Splitting strategy | Random | Random | Random | Scaffold | Scaffold | Scaffold |
AttentiveFP | 0.853 (0.060) | 2.030 (0.420) | 0.650 (0.030) | 0.877 (0.029) | 2.073 (0.183) | 0.721 (0.001) |
FragGAT | 0.878 (0.124) | 1.538 (0.640) | 0.645 (0.042) | 0.884 (0.041) | 2.065 (0.201) | 0.750 (0.013) |
MPNN | 1.167 (0.430) | 2.185 (0.952) | 0.672 (0.051) | 1.541 (0.630) | 2.430 (0.821) | 0.730 (0.063) |
DMPNN | 0.980 (0.258) | 2.177 (0.914) | 0.653 (0.046) | 1.050 (0.008) | 2.182 (0.183) | 0.683 (0.016) |
CMPNN | 0.789 (0.112) | 2.007 (0.442) | 0.614 (0.029) | 0.845 (0.039) | 1.833 (0.580) | 0.658 (0.029) |
CoMPT | 0.774 (0.058) | 1.855 (0.578) | 0.592 (0.048) | 0.915 (0.042) | 1.959 (0.808) | 0.646 (0.028) |
GROVERbase | 0.888 (0.116) | 1.592 (0.072) | 0.660 (0.061) | 1.185 (0.160) | 2.001 (0.081) | 0.817 (0.008) |
GROVERlarge | 0.831 (0.120) | 1.544 (0.397) | 0.643 (0.030) | 1.098 (0.178) | 1.987 (0.072) | 0.823 (0.010) |
PharmHGT | 0.680 (0.137) | 1.266 (0.239) | 0.583 (0.026) | 0.839 (0.049) | 1.689 (0.516) | 0.638 (0.040) |
