Bound ligand falls out of the pocket during ATMMinimization

[igashov]

Hello,

I am following the ATM tutorial and trying to compute a ddG between two compounds binding to PGK1. I am using the crystal structure of PGK1 with the bound compound and want to compare it to a similar ligand from another crystal structure.

I noticed that after minimization both ligands change their positions: the bound ligand leaves its pocket and sticks in another cavity, and the second ligands seems to follow the same movement in the solvent.

Before minimization:

After minimization:

I also noticed that if I turn off com_distance_restraint in the ATMMinimization the ligands don't move away any more. But in this case it happens during equilibration because it also has the same argument and I assume I am not actually supposed to turn it off for ATM...

I wanted to ask if anyone ever noticed similar things and if there are any tips on how to deal with this issue. I am attaching the input files for the protein and both ligands + solvated complexes before and after minimization.

pgk1_ATM_inputs.zip

Thanks!

[mb2055]

Hi,

I can confirm that I can recreate this issue locally.
The issue is in the logic that calculates the distances used for the COM restraints - they appear to always be set to pull the ligand towards the centre of mass of the protein regardless of its original location. So if the ligand is in a pocket 10Å from the centre of mass of the protein, the restraints will always pull it to 0Å from the centre of mass, which is clearly incorrect.

This shouldn't be too difficult to fix, its just a case of calculating the correct distances and passing them to openMM.

Working on a fix now, I'll update this thread once its done.

[igashov]

Thanks a lot!