2) Showed how CANs and the SL follow naturally from introducing constraints into the discriminator of GANs; 3)

We thank all the reviewers for their careful reviews. None of these follow trivially from Xu et al. In molecule generation, validity is encouraged using the reward-based approach of MolGANs, not the SL. We will clarify this in the discussion. Thank you, we were not aware of this link.