ABFE on Thombin system problems

[nividic]

Hi there,

We are trying to validate ABFE calculations by using Yank. We are interested to simulate a particular series of Thrombin inhibitors. The series is composed of weak binders however, the YANK simulations are reporting that mostly of the ligands are non-binders. I have attached one of these systems providing all the Yank required files including the Yank setting file “experiments.yaml”. The attached ligand system is the ligand 3c related to the paper

https://pubs.acs.org/doi/10.1021/jm9016416

The protein and ligand were prepared accordingly to this other paper:

https://pubs.acs.org/doi/10.1021/acs.jpcb.6b03296

We wonder if there is something wrong in our Yank setting and why the results are far from the experimental value.

In addition, in order to run these simulations we are using Yank restarting and I’m getting the following sort of warning:

DEBUG - yank.multistate.multistatereporter - Initial checkpoint file automatically chosen as /tmp/tmp0xlrx8cz/Yank ABFE_eramxedu/experiments/solvent1_checkpoint.nc 2019-05-15 06:03:15,880 - DEBUG - yank.multistate.multistatereporter - analysis_particle_indices != on-file analysis_particle_indices!Using on file analysis indices of [ 0 1 2 ... 30214 30215 30216]

however, the simulations seem to restart from the last Yank iteration, and I do not know if this is something to be worried or not. We are using the Yank version 0.23.7, openmm 7.2.2 and openmmtools 0.15.0. Please let us know if you need further info

Yank_Test_Case_Thrombin.zip

[andrrizzi]

Hi @nividic . That warning is not to be worried about. This could be a science issue, of course, but, here are a few things to check.

• solvent_dsl: resname HOH 'Cl-' 'Na+': I'm not 100% sure this is correct MDTraj DSL syntax. MDTraj usually raises an exception in this case, but I'd double-check that this selects the atoms you intend.
• splitting: 'V R R R O R R R V': I think our current best practice is to use V R O R V.
• number_of_iterations: 1000: This number of iteration may be low, so the results might simply be too far from convergence.
• type: PeriodicTorsionBoresch: For Boresch restraints, I'd suggest to manually specify the restrained atoms with something like

  restraint:
  type: PeriodicTorsionBoresch
  restrained_receptor_atoms: [atom1_index, atom2_index, atom3_index]
  restrained_ligand_atoms: [atom4_index, atom5_index, atom6_index]

  However, if you're restraining the wrong binding mode, I think this restraint will slow down convergence a lot, and you'll probably get very weird results after 1000 iterations so other types of restraints might be better.

[jchodera]

@nividic: Thanks so much for posting this!

Could you also upload a tarball of the whole protein:ligands series you were testing if you have it? It will be easier to make sure we test on multiple ligands from this series in case there is an offset.

I'll do some testing myself. I'll echo @andrrizzi's sentiment that 1000 iterations is too short to get anything useful from one of these calculations, but we should be able to come up with a robust protocol for you via local testing.

Thanks again!

[jchodera]

@nividic: Just following up on this: Can you upload a tarball of the whole protein:ligand series you were testing?

[nividic]

@jchodera I'm very sorry about my late reply but for some reasons I was not getting GitHub notifications. I've attached the posed Thrombin 3 series as described in the paper:

https://pubs.acs.org/doi/10.1021/jm9016416

I'm providing the protein as prepared .pdb file and the ligands as .oeb file. Is something that you can work with?

Thank you so much.

Thrombin3Series.zip

[jchodera]

Yes, thank you!