Table 4 Top-k accuracy \((\%)\) after re-ranking on USPTO-full
From: RetroRanker: leveraging reaction changes to improve retrosynthesis prediction through re-ranking
Models | Top-1 | Top-3 | Top-5 | Top-10 |
|---|---|---|---|---|
RetroSim | 32.8 | – | – | 74.1 |
GLN | 39.3 | – | – | 63.7 |
LocalRetro | 39.1 | 53.3 | 58.4 | 63.7 |
AT | 47.6 | 62.4 | 66.7 | 70.7 |
AT+RetroRanker\(^{*}\) | 48.0 (+0.4) | 64.1 (+1.7) | 68.5 (+1.8) | 71.7 (+1.0) |
AT+RetroRanker\(^{\dagger }\) | 48.8 (+1.2) | 64.7 (+2.3) | 68.8 (+2.1) | 71.7 (+1.0) |
R-SMILES | 48.9 | 66.5 | 71.8 | 76.8 |
R-SMILES+RetroRanker\(^{\dagger }\) | 49.0 (+0.1) | 67.2 (+0.7) | 72.6 (+0.8) | 77.3 (+0.5) |