Table 2 Top-k accuracy of rxn-ebm and RetroRanker over RetroXpert and GLN
From: RetroRanker: leveraging reaction changes to improve retrosynthesis prediction through re-ranking
Rank | RetroXpert | GLN | ||||
|---|---|---|---|---|---|---|
Original | rxn-ebm | RetroRanker* | Original | rxn-ebm | RetroRanker\(^{\dag }\) | |
Top-1 | 45.8 ± 0.3 | 42.7 ± 0.3 | 47.3 ± 0.7 (+4.6\(^{\ddag }\)) | 51.7 ± 0.3 | 52.3 ± 0.0 | 52.1 ± 0.5 (-0.2) |
Top-3 | 59.2 ± 0.3 | 62.0 ± 0.2 | 64.4 ± 0.7 (+2.4) | 67.8 ± 0.4 | 74.9 ± 0.3 | 74.9 ± 0.2 (+0.0) |
Top-5 | 63.0 ± 0.6 | 67.6 ± 0.1 | 70.3 ± 0.2 (+2.7) | 75.1 ± 0.3 | 82.0 ± 0.2 | 82.7 ± 0.2 (+0.7) |
Top-10 | 66.9 ± 0.3 | 73.0 ± 0.3 | 75.7 ± 0.2 (+2.7) | 83.2 ± 0.1 | 88.0 ± 0.0 | 89.3 ± 0.2 (+1.3) |
Top-20 | 69.9 ± 0.6 | 75.9 ± 0.1 | 77.1 ± 0.3 (+1.2) | 88.9 ± 0.1 | 91.4 ± 0.1 | 92.1 ± 0.2 (+0.7) |
Top-50 | 73.0 ± 0.7 | 77.3 ± 0.2 | 77.3 ± 0.3 (+0.0) | 92.4 ± 0.1 | 93.0 ± 0.1 | 93.2 ± 0.1 (+0.2) |