Skip to main content

Table 4 Comparison of model accuracy on single reactant reaction dataset using ECFP and MACCS keys

From: Substructure-based neural machine translation for retrosynthetic prediction

  MACCS ECFP (Radius, nBits)
1,1024 1,2048 2,1024 2,2048
Encoder 126 1024 2016 1024 2048
Decoder 126 1024 2027 1024 2048
Ave. length 74 56 56 74 74
  Accuracy
\(T_{c} = 1.0\) 29.0% 35.6% 37.6% 9.1% 10.1%
\(T_{c} \ge 0.85\) 57.7% 50.7% 52.0% 28.4% 30.0%
\(\overline{T_{c}}\) 0.84 0.80 0.80 0.66 0.66