2022-10-18 Protein-ligand benchmarks meeting notes

MCSS Lomap

• RG: currently working an alternate MCS approach for getting ligands out of RDKit; appears to give saner results than Lomap

  • JC : is this intended to take already-posed ligands?

    • or generating all possible atom mappings and score them? Using geometry?

  • RG: think API can support a generator of mappings; on scorers, not tweaking approach yet

    • this is something that F@H would be useful for us to explore

• JC : would be useful for JS to get familiar with API for atom mappings

• RG : Clara Crist has a paper out for atom mappings that indicates it’s not just a 2d problem; that inspired approach to treating it as a 3d problem

• JC : would be worthwhile coordinating on this? we can take advantage of our collective expertise here and avoid duplicating effort

• IA : would be value in this as a separate call

• DD: gufe#35 just merged, features abstract base class for AtomMapping: https://github.com/OpenFreeEnergy/gufe/pull/35/files

• RG : we also want to be able to handle protein mutation atom mappings with the same API, so collaborating here would be useful

• DD : we’ll use next week’s slot for a detailed discussion on atom mappings, opportunities to coordinate effort

  • this is also an opportunity to establish buy-in on gufe data model from others

• RG : on network generation, Mary Pittman has been jumping into network planning based on atom mapping

  • JC : could have separate chats for network planning, atom mapping

protein-ligand-benchmark

• IA : can’t generate atom mappings properly due to squashed rings

  • Melissa is currently redocking; so far appears to address #81

  • DD : we’ll await her PR, then IA and DD will review

• IA : on #68, won’t touch until we have new docked structures

• IA : on #76: WIP, could use feedback from JC

• IA : on #69, blocked by #81; can use Mapping.to_dict for serialization approach now that they are GufeTokenizables

• DD: Including data in the python package, now that we don’t need/use LFS.

• IP : in the future will be contributing data to this repo from Relay; what is the process for contributing?

  • JC : need a guide to contributing?

    • would probably want to re-dock using a standard procedure

    • homogeneity is useful, even if it currently requires a proprietary tool

• DD : would like to use the Relay submission as a forcing function for building a guide to contributing; also, would like someone else to dock these as a way to test the guide

• DD : may make sense to pursue a separate working group for protein-ligand-benchmark as a way to share the burden of advancing it, including preparation of new structures for those with access to Schrodinger tools

  • would spin out protein-ligand-benchmark development from this forum into one with identified stakeholders

  • would address concerns stakeholders (in particular OpenFE board) raised on developments without being looped in

Nonequilibrium cycling

• IP : we are able to execute nonequilibrium cycling from gufe objects

  • do have to do some processing on perses side to get into appropriate form for OpenEye

  • JC : can you create a gist where you pin down where/if key information is being lost in going from SDF → RDKIT → OpenFF → OpenEye

  • IP: This should be okay from the gufe side of things, we need to see if there is some information lost along the way.