which, as far as I can tell, in both cases overwrites xyz_prev with randomly initialised coordinates, whereas from table S8 and Algorithm 3 in the SI (as well as your comment in the first code snippet), it appears that the coordinates should be copied from the top template or from a previous recycling iteration. Is my understanding of this correct or is the current version the intended behaviour?
Thank you very much for making this amazing resource available to the community!
While delving into the code, I had a question about the flow of coordinate information from templates and previous recycling iterations in the model:
Specifically, in
data_loader.py, you have the below: https://github.com/baker-laboratory/RoseTTAFold-All-Atom/blob/b461e1201628ef5feb59b8f860002fe751c3fc2d/rf2aa/data/data_loader.py#L120-L124and similarly in
recycling.py: https://github.com/baker-laboratory/RoseTTAFold-All-Atom/blob/b461e1201628ef5feb59b8f860002fe751c3fc2d/rf2aa/training/recycling.py#L58-L63which, as far as I can tell, in both cases overwrites
xyz_prevwith randomly initialised coordinates, whereas from table S8 and Algorithm 3 in the SI (as well as your comment in the first code snippet), it appears that the coordinates should be copied from the top template or from a previous recycling iteration. Is my understanding of this correct or is the current version the intended behaviour?