jianlin-cheng/MPBind ? reverse-engineered prompt

Build me a Python tool that can take a folder of protein structure files and predict binding sites on them, then save the results back into PDB files so I can open them in PyMOL or ChimeraX and see the scores colored on the structure.

I want it to work for both regular PDB inputs and AlphaFold style structures, and I’d like a simple command line flow for running inference, choosing which binding types to predict, and picking a model version. If training support is already part of this project, keep that working too, so I can retrain on my own data later.

Please make the setup straightforward, use the provided weights and example data if possible, and keep the outputs organized in a clear results folder under whatever input folder I give it. If anything needs current docs or model downloads, look them up online if needed.