Skip to main content
Fig. 1 | Journal of Cheminformatics

Fig. 1

From: GEN: highly efficient SMILES explorer using autodidactic generative examination networks

Fig. 1

Tested architecture for SMILES generation. Architecture with two consecutive biLSTM layers used for deep-generative models for SMILES generation. a Original architecture with two consecutive LSTM layers, followed by a Dense output layer to predict the next character. b Modified architecture with two consecutive bidirectional LSTM layers. c Advanced architecture with one embedding biLSTM layers followed by multiple parallel bidirectional encoding layers and a merging layer (concatenated, averaged or learnable average). d Advanced architecture using parallel-concatenated architectures with multiplication of embedding and encoding layers. These layers are merge by concatenation, averaging or learnable weighted average

Back to article page