Fig. 2
From: CNN-based two-branch multi-scale feature extraction network for retrosynthesis prediction
The overview of CNN-TMN. The input of CNN-TMN is chemical molecules, while the output is the predicted label L. Firstly, the SMILES and ECFP are canonicalized and extracted using RDKit. Then, these two descriptors are then fed into two different branches of the feature extraction network to extract their multi-scale features respectively. Finally, the concatenated features are fed into the full connection and softmax layer to get the final prediction result