Is it possible to use this strong baseline on binding affinity prediction?

[JosieHong]

Nice work!

I am interested in binding affinity prediction on virtual screening. Can I utilize this model to do that?

Thanks, Josie

[alexanderrich]

Hi Josie, you could definitely use a modified version of this code to predict binding affinity. A couple of things to note:

• You would want to use gnina's CNNaffinity output to compare affinity across ligands, rather than the CNNscore output that is intended for ranking poses within a ligand.
• The code here isn't really optimized for speed. If you want to screen large libraries, you could consider using something like Uni-dock that can take advantage of a GPU to greatly speed up pose sampling (gnina only uses the GPU for scoring poses after they are sampled), and then use gnina to rank the generated poses.

[JosieHong]

It is very helpful! Thank you so much.