Skip to main content
Fig. 4 | Journal of Cheminformatics

Fig. 4

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

Fig. 4

Global novelty analysis. For all sets of 2M generated compounds, the dataset has been split into 10k time points. a Plot showing the percentage of molecules at every time point t. b Cumulated number of unique molecules generated during the process. The final values for the three tested architectures are 1470,543 (73.5% efficiency) for LSTM–LSTM, 1566,535 (78.3% efficiency) for biLSTM–biLSTM and 1602,018 (80.1% efficiency) for biLSTM–biLSTM with 4 parallel-concatenated encoding layers

Back to article page