Skip to main content
Fig. 12 | Journal of Cheminformatics

Fig. 12

From: SMILES-based deep generative scaffold decorator for de-novo drug design

Fig. 12

Process how the scaffold generator and the decorator sample new SMILES. The Scaffold Generator (top) samples the tokens one at a time. The decorator (bottom), on the other hand, encodes the scaffold using a bidirectional RNN. Each output state and hidden state of the encoder is summed up. The output hidden states from the encoder are input directly to the decoder. Legend: squares are instances of RNN as described in Fig. 11; gray lines hidden states in each of the sampling steps; \(\oplus\) symbolizes the position-wise sum of all tensors

Back to article page