Table 3 A comparison of reported top-1 accuracies of retrosynthesis prediction models without additional reaction classes.

From: Retrosynthetic reaction pathway prediction through neural machine translation of atomic environments

Model

Top-1(%)

Non-transformer

 Coley et al., similarity-based, 201772

32.8

 Segler et al.,–rep. by Lin, Neuralsyma, 20206,33,73

47.8

 Dai et al., Graph Logic Networka, 201973

39.3

 Liu et al.,–rep. by Lin, LSTM-based, 202016,33

46.9

 Genheden et al., AiZynthfinder, ANN + MCTSa, 202014,49

43–72

Transformer-based

 Zheng et al., SCROP, 202034

41.5

 Wang et al., RetroPrime, 202148

44.1

 Tetko et al., Augmented Transformer, 202035

46.2

 Lin et al., AutoSynRoute, Transformer + MCTS, 202033

54.1

 RetroTRAE

58.3

 RetroTRAE (with SM and DM)

61.6

  1. The results are based on either filtered MIT-full46,47 or MIT-fully atom mapped15 reaction datasets.
  2. aReaction templates were used.