search

gcorso / DiffDock

Implementation of DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking
https://arxiv.org/abs/2210.01776
MIT License
1.1k stars 266 forks source link

If you want to benchmark the virtual-screening power #4

Closed UnixJunkie closed 2 years ago

UnixJunkie commented 2 years ago

LIT-PCBA: An Unbiased Data Set for Machine Learning and Virtual Screening https://pubs.acs.org/doi/10.1021/acs.jcim.0c00155

Dataset: https://drugdesign.unistra.fr/LIT-PCBA/

gcorso commented 2 years ago

Why would this dataset be more suited than PDBBind to evaluate the capacity to perform docking?

UnixJunkie commented 2 years ago

As far as I know, PDBBind is about binding energy or so. It does not tell you anything about the virtual screening power of a method.

gcorso commented 2 years ago

Well, PDBBind also gives binding energy, but we are not even interested in that at the moment. We are evaluating the quality of our method to dock the ligand in the correct pose. How to use this to do virtual screening is a whole other research question that has been studied for decades with existing effective solutions, but that we have not yet tackled directly.