Skip to main content
. 2023 Oct 3;14:6155. doi: 10.1038/s41467-023-41698-5

Table 1.

Performance of our RetroExplainer and the state-of-the-art methods on USPTO-50K benchmarks

Model Top-k accuracy (%)
Reaction class known Reaction class unknown
k = 1 3 5 10 1 3 5 10
Fingerprint-based
RetroSim41 52.9 73.8 81.2 88.1 37.3 54.7 63.3 74.1
NeuralSym8 55.3 76.0 81.4 85.1 44.4 65.3 72.4 78.9
Sequence-based
SCROP59 59.0 74.8 78.1 81.1 43.7 60.0 65.2 68.7
LV-Transformer23 - - - - 40.5 65.1 72.8 79.4
AutoSynRoute60 - - - - 43.1 64.6 71.8 78.7
TiedTransformer61 - - - - 47.1 67.1 73.1 76.3
MolBART62 - - - - 55.6 - 74.2 80.9
Retroformer63 64.0 82.5 86.7 90.2 53.2 71.7 76.6 82.1
RetroPrime64 64.8 81.6 85.0 86.9 51.4 70.8 74.0 76.1
R-SMILES65 - - - 56.3 79.2 86.2 91.0
DualTF46 65.7 81.9 84.7 85.9 53.6 70.7 74.6 77.0
Graph-based
GLN36 64.2 79.1 85.2 90.0 52.5 69.0 75.6 83.7
G2Gs17 61.0 81.3 86.0 88.7 48.9 67.6 72.5 75.5
G2GT18 - - - - 54.1 69.9 74.5 77.7
GTA16 - - - - 51.1 67.6 73.8 80.1
GraphRetro33 63.9 81.5 85.2 88.1 53.7 68.3 72.2 75.5
Graph2SMILES39 - - - - 52.9 66.5 70.0 72.9
RetroXpert32 62.1 75.8 78.5 80.9 50.4 61.1 62.3 63.4
GET38 57.4 71.3 74.8 77.4 44.9 58.8 62.4 65.9
LocalRetro57 63.9 86.8 92.4 96.0 53.4 77.5 85.9 92.4
RetroExplainer (Ours) 66.8 88.0 92.5 95.8 57.7 79.2 84.8 91.4

The performance regarding existing methods is derived from their references. The best-performing results are marked in bold.