Interpreting Deg_adj_list

Hi,

I am exploring bbbp dataset and I came across this Deg_adj_list in the convmol object. I don’t have much background in deep learning. Please can someone help me understand what deg_adj_list corresponds to or some article which explains what is supposed to be ? Thank you.

If I recall, it’s a sorting of the neighbors of an atom according to the degree of the bond (single/double). I’m going off memory so I could be inaccurate here

1 Like

I believe that “degree” here refers to the number of other atoms that an atom is bonded to. A graph convolution involves multiplying by matrices whose size matches the number of bonds connected to an atom, so it creates adjacency lists that are split up by the number of bonds. I haven’t worked on this code in quite a while either, so my recollection also might not be entirely accurate.

2 Likes

thank you @peastman and @bharath for the responses, I am able to understand that it has to do something with the connectivity but I am still confused. Is there any documentation I can refer to so that I can use the deg_adj_list info in my model? Thank you.

Check out https://deepchem.readthedocs.io/en/latest/api_reference/data.html#deepchem.feat.mol_graphs.ConvMol. As a warning up front, these classes aren’t part of DeepChem’s stable API, so we may make breaking changes here in future releases (see the notes in the docs). Good luck and happy hacking!