Odd data in PDBBind2016

[hnisonoff]

I wanted to flag some oddities with the PDBBind2016 dataset. I've tried to recompute the RMSDs and have noticed a very large fraction do not match the data. One particularly odd example I found was in 5c28 where the docked ligand is a different molecule from the crystal ligand. Is there by any chance a cleaner version of the PDBBind docked dataset that could be used?

[francoep]

The data we utilized is downloaded directly from the PDB at the time of the data creation, we only took the affinity numbers from the PDBbind data and were matching PDB+ligname from the PDBbind to what was identified via pocketome.

Could you give more information about this example (Pocket & specific files)?

[francoep]

Or did I misunderstand, and you meant exactly the PDBbind2016 data?

[francoep]

Additionally, looking at the 5c28 example you mentioned, the molecules are the same. Could you provide more examples of faulty data, and also describe how you are trying to re-compute the RMSDs?

[hnisonoff]

I am describing the 5c28 example from PDBbind2016.tar.gz. I just double check by redownloading everything.

Here is an image showing what happens when I load the docked molecule and the one labeled as ligand. As you can see they are different.

[hnisonoff]

Here is the directory. Files are 5c28_docked.sdf and 5c28_ligand.sdf
5c28.zip

[dkoes]

Doesn't look different to me. Looks like the same molecule rotated 180 degrees.

[hnisonoff]

Sorry you are right. I apologize for the inconvenience. For RMSDs I was using rdkit CalcRMS and spyrmsd. I'll go back and fill this out with more detail. Sorry again for the incorrect bug report.

[francoep]

For the RMSDs that we reported, we used obrms to calculate them (comes when you install openbabel).

[dkoes]

CalcRMS does not do symmetry correction. spyrmsd is suppose to. I'd be interested in seeing examples where obrms and sprmsd differ.

[hnisonoff]

I'll try to dig up examples:

For CalcRMS, the documentation says: "Note: This function will attempt to align all permutations of matching atom orders in both molecules"

Doesn't this imply it does symmetry correction?

[dkoes]

Huh, you're right - that's what the documentation says. Not sure why there is also GetBestRMS then. Maybe historically it CalcRMS didn't do that?

[drewnutt]

GetBestRMS calculates RMSD and then aligns them in space.
CalcRMS calculates the RMSD without moving either molecule.

[JonasLi-19]

I suppose that 'obrms' also calculates the RMSD without moving either molecule?

[dkoes]

Yes, unless -m is passed.