Fig. 9From: Molecular de-novo design through deep reinforcement learningA (small) change of mind. The conditional probability distributions when the DRD2 test set active ‘COc1ccccc1N1CCN(CCCCNC(=O)c2ccccc2I)CC1’ is generated by the Prior and an Agent trained using the DRD2 activity model, for the case where all actives used to build the activity model have been removed from the Prior. E = EOSBack to article page