Closed hmms117 closed 2 years ago
Hi!
EquiBind does (currently) not provide a confidence score or something similar to a docking score which could be used for ranking. If the intersection loss is unusually high it would indicate that there are intersections with the receptor and the conformation is likely not the best. However, this could also be seen in a visual inspection.
Thanks for the swift response! Ok, so for testing EquiBind for ranking, following EquiBind, we could try to run the (slower) tools to obtain a docking score.
With visual inspection, we see consistent clashes in our own test examples (PDB and ligands clashes). At inference time, are the PDB coordinates also adjusted? This happens in some docking tools. Do we need to save the PDB coordinates along with the ligand coordinates in inference_from_files
? We are testing on single chain PDB files and a ligand in .sdf
format.
Again, thanks for developing the method!
for testing EquiBind for ranking, following EquiBind, we could try to run the (slower) tools to obtain a docking score.: correct!
The coordinates of the receptor are not adjusted if that is what you are asking. The receptor is treated as a rigid body.
The receptor files also need to be provided next to the ligand files. It does not matter whether they are provided as .pdb or .sdf. I am not sure whether that answers the question. Please let me know if unclarity remains!
Thanks! It makes good sense - we will try it out! :)
Hi,
First of all, thanks for developing this method - it is really something new.
When using the
inference_from_files
mode to infer various ligand conformations on several protein targets, how does the tool report some measure of inference quality"? Do you have some measure of conformation fit to the target? I can seeintersection_losses_untuned
being reported - can it be used?Many thanks!