Retrosynthesis on USPTO-50K Reaction types given as prior (test)
66.5Top-1 AccuracyGLN w/ DMP
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GLN w/ DMPPre-training=DMP2021.06 | 66.5 | 81.2 | 86.6 | 90.5 | 92.8 | 93.5 | |
| GLN w/ DMPFramework=GLN, Pre-training=DMP2021.06 | 66.5 | 81.2 | 86.6 | 90.5 | 92.8 | 93.5 | |
| GLN2021.06 | 64.2 | 79.1 | 85.2 | 90 | 92.3 | 93.2 | |
| GMTBackbone=GLN, Pooling Method=Graph Multiset Transformer (GMT), Reaction Class Provided=true2021.02 | 64.17 | 79.61 | 85.32 | 89.97 | 92.31 | 93.25 | |
| MinCutPoolBackbone=GLN, Pooling Method=MinCutPool, Reaction Class Provided=true2021.02 | 63.91 | 79.19 | 84.76 | 89.69 | 92.13 | 93.23 | |
| GLNBackbone=GLN, Pooling Method=Average Pooling, Reaction Class Provided=true2021.02 | 63.53 | 78.27 | 84.32 | 89.51 | 92.17 | 93.17 | |
| GLN w/ DeeperGCNGNN architecture=DeeperGCN2021.06 | 63.2 | 79.2 | 85 | 90 | 92 | 93.1 | |
| GLN w/ DeeperGCNFramework=GLN, Backbone=DeeperGCN2021.06 | 63.2 | 79.2 | 85 | 90 | 92 | 93.1 | |
| GLN2021.06 | 63.2 | 77.5 | 83.4 | 89.1 | 92.1 | 93.2 | |
| DMP fusionPre-training=DMP2021.06 | 57.5 | 75.5 | 80.2 | 83.1 | 84.2 | 85.1 | |
| DMP fusionPre-training=DMP2021.06 | 57.5 | 75.5 | 80.2 | 83.1 | 84.2 | 85.1 | |
| ChemBERTa fusionPre-training=ChemBERTa2021.06 | 56.4 | 74.7 | 78.9 | 81.8 | 83.3 | 84.5 | |
| ChemBERTa fusionBase Model=ChemBERTa2021.06 | 56.4 | 74.7 | 78.9 | 81.8 | 83.3 | 84.5 | |
| NeuralSym2021.06 | 55.3 | 76 | 81.4 | 85.1 | 86.5 | 86.9 | |
| TransformerModel type=Standard Transformer2021.06 | 54.2 | 73.6 | 78.3 | 81.3 | 83.1 | 84.3 | |
| Transformer2021.06 | 54.2 | 73.6 | 78.3 | 81.3 | 83.1 | 84.3 | |
| RetroSim2021.06 | 52.9 | 73.8 | 81.2 | 88.1 | 91.8 | 92.9 |