Table 1 Top-k accuracies of RSGPT model and baselines on USPTO-50k60 dataset with reaction class unknown
Model | Reaction class unknown | |||
|---|---|---|---|---|
Top-1 | Top-3 | Top-5 | Top-10 | |
Template-based | ||||
RetroSim61 | 37.3 | 54.7 | 63.6 | 74.1 |
NeuralSym62 | 44.4 | 65.3 | 72.4 | 78.9 |
GLN8 | 52.5 | 69.0 | 75.6 | 83.7 |
LocalRetro52 | 53.4 | 77.5 | 85.9 | 92.4 |
RetroComposer9 | 54.5 | 77.2 | 83.2 | 87.7 |
Semi-template-based | ||||
G2G63 | 48.9 | 67.6 | 72.5 | 75.5 |
RetroXpert51 | 50.4 | 61.1 | 62.3 | 63.4 |
RetroPrime64 | 51.4 | 70.8 | 74.0 | 76.1 |
G2Retro65 | 54.1 | 74.1 | 81.2 | 86.7 |
SemiRetro12 | 54.9 | 75.3 | 80.4 | 84.1 |
Graph2Edits13 | 55.1 | 77.3 | 83.4 | 89.4 |
Template-free | ||||
SCROP17 | 43.7 | 60.0 | 65.2 | 68.7 |
MEGAN66 | 48.1 | 70.7 | 78.4 | 86.1 |
Graph2SMILES18 | 52.9 | 66.5 | 70.0 | 72.9 |
R-SMILES43 | 56.3 | 79.2 | 86.2 | 91.0 |
NAG2G10 | 55.1 | 76.9 | 83.4 | 89.9 |
EditRetro45 (×20) a | 60.8 | 80.6 | 86.0 | 90.3 |
RSGPT | 63.4 | 84.2 | 89.2 | 93.0 |
RSGPT (×20) a | 77.0 | 90.9 | 94.3 | 96.7 |
