Table 4 Ablation study on the effectiveness of using different molecular descriptors. ECFP and SMILES indicate that CNN-TMN only uses ECFP or SMILES descriptor for retrosynthesis prediction
From: CNN-based two-branch multi-scale feature extraction network for retrosynthesis prediction
ECFP | SMILES | Top-1 | Top-3 | Top-5 | Top-10 |
|---|---|---|---|---|---|
\(\checkmark\) | – | 47.2 | 67.6 | 73.5 | 80.0 |
– | \(\checkmark\) | 53.8 | 73.1 | 78.7 | 84.8 |
\(\checkmark\) | \(\checkmark\) | 61.1 | 79.1 | 83.9 | 87.7 |