Table 3 Performance comparison on property prediction of small molecules (classification tasks)
From: Force field-inspired molecular representation learning for property prediction
Metric | ROC-AUC ↑ | |||||
|---|---|---|---|---|---|---|
Dataset | BACE | BBBP | HIV | Tox21 | SIDER | ClinTox |
SVM | 0.811 (0.054) | 0.829 (0.060) | 0.627 (0.009) | 0.822 (0.006) | 0.682 (0.013) | 0.669 (0.092) |
RF | 0.815 (0.049) | 0.790 (0.062) | 0.645 (0.015) | 0.769 (0.015) | 0.684 (0.009) | 0.713 (0.056) |
GATv2 | 0.843 (0.035) | 0.893 (0.021) | 0.818 (0.012) | 0.840 (0.026) | 0.618 (0.036) | 0.694 (0.146) |
GIN | 0.850 (0.031) | 0.890 (0.007) | 0.786 (0.031) | 0.824 (0.015) | 0.619 (0.009) | 0.753 (0.132) |
GCN | 0.829 (0.037) | 0.895 (0.003) | 0.752 (0.020) | 0.788 (0.025) | 0.624 (0.019) | 0.615 (0.013) |
DMPNN | 0.878 (0.032) | 0.913 (0.026) | 0.816 (0.023) | 0.845 (0.015) | 0.646 (0.016) | 0.894 (0.027) |
DimeNet | 0.832 (0.023) | 0.822 (0.040) | 0.724 (0.016) | 0.758 (0.019) | 0.626 (0.008) | 0.738 (0.020) |
FFiNet | 0.891 (0.016) | 0.916 (0.012) | 0.828 (0.010) | 0.852 (0.009) | 0.656 (0.017) | 0.919 (0.021) |