Hi, thanks for your amazing work! After reading your Diffdock-PP paper and running the code, I am curious why you chose the VE-SDE for the score model. Have you tried other SDEs like VP and the one in the EDM paper? Will the schedule greatly affect the sample quality due to the different weights in the exploring and refinement stages?
Besides, I find a bug in the code for atom features. It seems you put the residue type at the end of atom features but use the front of atom features as the residue type in the atom embedding layer during the forward function.
Hi, thanks for your amazing work! After reading your Diffdock-PP paper and running the code, I am curious why you chose the VE-SDE for the score model. Have you tried other SDEs like VP and the one in the EDM paper? Will the schedule greatly affect the sample quality due to the different weights in the exploring and refinement stages?
Besides, I find a bug in the code for atom features. It seems you put the residue type at the end of atom features but use the front of atom features as the residue type in the atom embedding layer during the forward function.