Graph Classification on MoleculeNet (1:1:8 scaffold split test)
70.07Tox21 AccuracyMole-BERT
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Mole-BERTBackbone=GIN, Pre-training Strategy=Mole-BERT2026.06 | 70.07 | 59.72 | 52.58 | 55.52 | 61.05 | 57.79 | 68.44 | 60.07 | 4 | |
| PatchNet (AM+CP)Backbone=PatchNet, Pre-training Strategy=AM+CP2026.06 | 68.56 | 59.64 | 53.06 | 59.5 | 62.19 | 58.74 | 68.55 | 59.74 | 1.4 | |
| GraphCLBackbone=GIN, Pre-training Strategy=GraphCL2026.06 | 68.22 | 59.09 | 52.67 | 56.99 | 58.73 | 56.82 | 64.68 | 56.92 | 7.9 | |
| GraphMVPBackbone=GIN, Pre-training Strategy=GraphMVP2026.06 | 68.01 | 55.43 | 52.24 | 55.54 | 57.36 | 56.88 | 65.41 | 57.77 | 10.3 | |
| GPT-GNNBackbone=GIN, Pre-training Strategy=GPT-GNN2026.06 | 67.98 | 58.39 | 52.97 | 57.07 | 58.56 | 56.68 | 65.06 | 56.25 | 8.5 | |
| GIN (w/o pre)Backbone=GIN, Pre-training Strategy=None2026.06 | 67.9 | 58.39 | 52.14 | 56.43 | 58.53 | 56.58 | 58.57 | 55.84 | 12.5 | |
| PatchNet (w/o pre)Backbone=PatchNet, Pre-training Strategy=None2026.06 | 67.77 | 56.74 | 51.88 | 55.78 | 60.16 | 57.72 | 59.15 | 57.35 | 10.4 | |
| AM+CPBackbone=GIN, Pre-training Strategy=AM+CP2026.06 | 67.62 | 58.19 | 52.44 | 57.17 | 59.06 | 56.53 | 63.79 | 57.96 | 9.3 | |
| AttrMaskBackbone=GIN, Pre-training Strategy=Attribute Masking2026.06 | 67.17 | 59.33 | 52.21 | 56.69 | 58.58 | 57.34 | 63.65 | 57.27 | 9.3 | |
| 3D InfoMaxBackbone=GIN, Pre-training Strategy=3D InfoMax2026.06 | 67.05 | 58.22 | 52.58 | 54.56 | 59.85 | 56.65 | 67.64 | 58.66 | 9 | |
| ContextPredBackbone=GIN, Pre-training Strategy=Context Prediction2026.06 | 66.45 | 58.16 | 51.53 | 55.83 | 59.49 | 56.58 | 63.57 | 57.92 | 11.3 | |
| PatchNet (CP)Backbone=PatchNet, Pre-training Strategy=Context Prediction (CP)2026.06 | 66.22 | 59.56 | 53.06 | 58.6 | 61.79 | 54.11 | 55.01 | 57.55 | 7.9 | |
| PatchNet (AM)Backbone=PatchNet, Pre-training Strategy=Attribute Masking (AM)2026.06 | 64.84 | 56.95 | 52.29 | 57.26 | 60.65 | 54.27 | 60.32 | 57.2 | 11.1 |