Fig. 2. Performance of the best model on four different translation tasks during the first 20 000 training steps. The Sml2canSml* run was trained without the additional classification task of molecular properties. (a) Translation accuracy. (b) Mean performance on the lipophilicity regression task. (c) Mean performance on the Ames classification task. For (b) and (c), the translation model at the respective step was utilized to extract the molecular descriptors fed into an SVM to model both tasks.