Demultiplexing RepEx trajectories

[dlukauskis]

Hi, I'd like to use openmmtools's replica exchange code, but I can't see a way of actually demultiplexing the trajectories and having a look at them. Am I missing something? The docs don't address this point.

Thanks, Dom

[andrrizzi]

Hi @dlukauskis . The Repex facilities were ported from YANK, but unfortunately we forgot to carry over the utilities to handle the trajectories in the netcdf file. You can take a look at this function https://github.com/choderalab/yank/blob/master/Yank/analyze.py#L1036-L1202 .

[dlukauskis]

I gave the YANK code a try but I'm getting a metadata-related error:

  File "sim_extract_trj.py", line 32, in <module>
    replica_index=0)
  File "/home/dom/Projects/Playground/SAMS/demux_test/extract_trajectory.py", line 54, in extract_trajectory
    reference_system = mmtools.utils.deserialize(metadata['reference_state']).system
KeyError: 'reference_state'

Attaching a directory with the exact code I used - demux_test.zip

[andrrizzi]

Yes, that's because YANK stores some metadata with the reference system and the topology. You will have to modify those bits and provide your own OpenMM Topology object.

[simomarsili]

This should solve the problem (until extract_trajectory will be ported to opemmtools):

https://github.com/simomarsili/mmdemux

[andrrizzi]

Thanks for sharing @simomarsili !

[tanmoy7989]

@andrrizzi, Is demultiplexing now included in MultistateAnalyzer? I'm still fairly new to the openmm environment (and openmmtools as well), so didn't go through the MultiStateAnalyzer methods in detail before asking.