How to DIY my own CNN scoring function model into this Vina python-binding?

[JonasLi-19]

Dear developer, I desperately need a method to use my own DL model scoring function (which is a CNN model constructed by Tensorflow2) to replace the "vina" scoring function to dock ligands. It is important that During my whole project, my model is always changing (I am doing some research to evaluate the docking power of the scoring function during a special long process), so every turn I get the new weights file of the CNN, I need to let it dock the dataset and evaluate the output. I will be very grateful if you can teach me how to accomplish this task, or what I should learn first. Thank you in advance.

[diogomart]

You'd have to study the source code and modify it for your purposes. Gnina does what you describe, see https://github.com/gnina/gnina