Project Plan: Common workflow conversion via Interchange

User story - A potential user has a simulation setup/parameterization workflow in a different simulation ecosystem. Their workflow is based on common usage of that ecosystem. That ecosystem is not particularly easy to use or extend, so they’re interested in migrating to OpenMM and possibly starting to use OpenFF tools. They navigate to a source of information that says “OpenFF validated that outputs from your workflow can be converted successfully to OpenMM”, and proceeds with the migration.

End goal - A one time artifact (possibly blog post, optionally with a repo) stating whether Interchange can get various ecosystem tutorials to get identical energy evaluations in OpenMM, and documenting the severity/details of disagreement if not. 

• Week 1 in progress: Select condensed phase system setup tutorials (AMBER+GROMACS+ optionally others)

  • Ideally 5+ (it’s fine if there are many simple ones)

  • Exclude tutorials that need more than conda-installable deps

  • Aim for a diversity of inputs (small mols, protein, nucleic acids, lipids)

  • Optionally, include a collection of “won’t even bother” tutorials (vsites, metals, alternative nonbonded)

  • Goal: Present the tutorials selected and what kinds of coverage they add to study

• Week 2 not started : Select a specific MD config, parameter, and coordinate file from each tutorial.

  • The files selected don’t need to be the minimized/post-simulation coordinates, but they should include all chemical components in the final system.

  • Filter out ones that don’t work at all

  • Optionally, if components from tutorial are output separately, try combining them using `+` operator

  • Goal: Present the run config, parameter, and coordinate files to be used for each tutorial, documenting any decisions made or deviations from the tutorial to get them.

• Week 3 not started : Evaluate energies for each tutorial’s files natively and in OpenMM via Interchange

  • Run in both the native engine and OpenMM, comparing energy components.

  • Optionally, fix “low-hanging fruit”

  • Optionally, do also convert from native to OpenMM using ParmEd

  • Goal: Present a table of energy values (native vs. OpenMM-via-Interchange) for the systems selected from each tutorial, ideally broken down by component

• Week 4 not started : Write up one-time study/blog post

  • Add Zenodo DOI in some form

  • Goal: A website PR adding a blog post presenting the above table and explaining how the study was run. 

Possible followups:

• Here’s how you would run these tutorials, substituting in OpenFF components, possibly looking at the quality of results in following analysis

• Here’s how these results look N months/years later

Not in scope:

• Sharing or maintaining the code that ran this study past the posting of results

• Updating this study past the posting of results (if there’s a good reason to rerun then that can be pitched as a new project)

• Turning this into a paper (though sticking a Zenodo DOI on the blog post could be in scope if there’s demand)