PXR user case

[jthorton]

OpenFE Failure Case Submission

Failure type

• Crash
• NaN result
• Large variation between repeats (~5 kcal/mol or more)
• Other:

Notes

Add any extra context that could help reproduce or understand the failure.

• Did you try to run the same transformation using a different OpenFE protocol (SepTop or plainMD)? No due to missing residues this system needs to be re-preped and run again.
• Did you try to run the same transformation using a different atom mapping method? No
• Did you try to run the same transformation using a different force field? No
• Where automatic protocol restarts enabled (by default this is true of openfe quickrun)? Yes
• Does the simulation crash immediately or after some progress? This is case-specific; see comment.
• Do repeats crash at the same point in the simulation? Only a single repeat provided by the user
• How were the input structures prepared? (e.g., docking, co-folding, etc. provide as much detail as possible) Docking into xtal structures.

Attachments

• Protocol Transformation JSON(s)
• Logs
• For NaN results the state.xml file stored in the nan-error-logs folder
• Environment YAML (conda list output)
• Hardware details

Environment details

• OpenMM accelerator details: (CPU / GPU, CUDA version, etc.)
  • Mix of NVIDA GPUs running CUDA 12.8
  • openfe-1.11.1
  • openmm - 8.2
• OS:: Linux

[jthorton]

rbfe_OCNT-2311573_OADMET-0006255
Crashes in the solvent leg with a reasonable mapping; the hybridisation change near the alchemical region could be the cause of the simulation instability, as it crashes after a few equilibration iterations.

rbfe_OADMET-0006097_OADMET-0006093
The complex leg crashes for this transformation due to missing residues in the xtal structure which are not modelled or capped, causing OpenMM to bond distant residues together and crash in the first equilibration iteration.

Screenshot showing uncapped residue in the input structure

rbfe_OADMET-0006193_OADMET-0006175
The atom mapping is empty for this transformation due to no spatial overlap, but the MST network planner returns this as an edge for the network when it probably should not.

This crashes before starting the simulation while trying to create hybrid topology with an Index error.

rbfe_OADMET-0006142_OCNT-2316690
This transformation has very low spatial overlap, resulting in a single mapped heavy atom, while the simulation does run in the solvent leg it has a large MBAR error (~2.3 kcal/mol) and pymbar prints many warnings about struggling to converge this also makes each iteration slower.

The complex leg fails but this could also be due to an issue with missing caps in the protein.

[hannahbaumann]

Thanks @jthorton ! Just the one comment, otherwise lgtm!

[hannahbaumann]

Is this supposed to be empty or does it have something missing?