one small bug，I hope it will be helpful

[Turningl]

Hello jk: when I test your data_structs.py code, I find a small bug in it, as follow: the original smiles: COc1cc(N2C(N)=C(C#N)C(c3ccccc3)C3=C2CC(C)(C)CC3=O)cc(OC)c1OC the transformer matrix: [51. 52. 54. 16. 20. 44. 6. 44. 44. 2. 19. 7. 16. 2. 19. 3. 15. 16. 2. 16. 0. 19. 3. 16. 2. 44. 8. 44. 44. 44. 44. 44. 8. 3. 16. 8. 15. 16. 7. 16. 16. 2. 16. 3. 2. 16. 3. 16. 16. 8. 15. 20. 3. 44. 44. 2. 20. 16. 3. 44. 6. 20. 16. 48.] the generative smiles: not_DBrD2high_QEDgood_SACOc1cc(N2C(N)=C(C#N)C(c3ccccc3)C3=C2CC(C)(C)CC3=O)cc(OC)c1OC the generative smiles include some boolean attributes, but the 'DRD2' attribute will be converted to 'DBrD2'，so the decode funciton has some problem, so I make a little change in it. I hope it will be helpful! As I step through your code, I seem to realize that the deocde part in your 2_generator_Transformer.py is a very important part, and I really want to know if you decode the generated csv file without errors in this part?

[jkwang93]

Thank you very much for your suggestion. We did not have this problem during the test, data_structs.py does require users to make changes according to their own needs. But the situation you encountered is indeed quite unexpected. We will confirm the bug and make adjustments later.

[Turningl]

thank you verr much for your reply. I would also like to ask you another question, I have to say that the effect of smiles gnerated by transformer model is good after I reproduce it, and that's where my problem lies. I know when you encode, you have programmed some boolean attributes (QED, SA, JNK3...) as 0 and 1, but the smiles generated by 2_generator_Transfomer.py the decode part don't seem to have these boolean attributes (QED, SA, JNK3...) , generated smiles are intact, why would these smiles not have any relevant attributes attached? I really want to know！

[Turningl]

emmm, I think I have found the answer myself. But antoher question is I find 4_train_agent_save_smile.py can't generative effectively molecular, I hope you can confirm the bug and make adjustments later

[jkwang93]

There should be no problem with this part of the code, you can email me if you have any specific problems. jikewang {at} whu {dot} edu {dot} cn

[Turningl]

I have emailed to you, please check.

[Turningl]

this problem hasn't been solved, so I'm curious to know what your changes are yet？

[bbjy]

I meet the same problem. The generated molecules by 3_train_middle_model_dm.py have unexpected attribute tokens, e.g. "high_QEDgood_SACc1cc(-n2cc(C(F)(F)F)nn2)cc(OS(=O)(=O)NCC2CC2)c1F".

Do you have solved this problem? Thank you!