alpha5 of recharge can now do wavefunction reconstruction and ESP output.
Looked over Espaloma, studied ML principles. Not sure that it’s mature in its current form, but principles should be applicable in a production package. Should be possible to rewrite in a more production-ready language. Did some experimentation with the same framework on AM1BCC-style functionality.
MT – Agree.
SB – As I understand it, the output could be immediately-usable MM parameters, which should be easily consumable/storable by the System object.
ML – Yes, I agree.
JW – Espaloma working session can help resolve some details of interface (eg, should System record entire ML model, or just outputs? in what format?)
SB – Espaloma models should be able to fit in memory pretty easily
JW – Should see if JC expects to be able to take a OpenFF System + espaloma model and differentiate the whole thing.
SB – Wouldn’t consider specifically “espaloma models” – Look at how to generally interface with GCNs.
JW – Agree
MT – Agree. Also want to think about timemachine the same way – Helpful for desciribing a sort of tool we want to interface with, but we shouldn’t hold ourselves to it as an exact interface/requirement.
(General) – SB is still working on role transition, but will generally be available during OpenFF working hours.
MT –
Lots of small things on many projects (CMILES refactor, etc)
Studying unit-bearing fields in pydantic (quantity of list vs. list of quantity vs. arrays) – There may be more options than we originally considered, will research further.
SB – We should have another working session on this, and come up with a pattern that we can use throughout the consortium.
MT – Big new wrinkle is arrays vs. floats storage/serialization format.
Switched toolkit CLI from argparse to click
Some system progress
JW –
Finished up hiring for scientific communicator
Lots of benchmarking work + user support, especially regarding benchmarking workflow.
PB –
Work on WBO interpolated parameters. Seeing some improvement compared to vanilla parameters.
JW – Maybe consider doing benchmarks using 1.3.0 benchmark set
Started doing submissions from genentech set. Filtered to simple mols (neutral, < 3 rotors) and open for review. Next set will be >=3 rotors
DH –
Transitioning to full position at Janssen in Jan. Will be doing lots of project work, OpenFF stuff will wind down. Will be overseeing new OpenFF postdoc.
Worked on druglike molecule/drugbank-PDB study (similar to Lim study, on a larger set). In addition to comparing to QM, can also compare to bound ligand conformations.
JW – For comparing to bound xtal poses, consider a scoring metric that ignores “correct” atom placement and only considers how well the predicted pose fits electron density.
Work on plbenchmarks repo – Happy with git lfs. Currently data is in a separate repo, but I may move it into main PLBenchmarks repo.
Getting ready for PLBenchmarks release (and maybe a name change)
Name suggestions?
openff-beryllium
openff-plbenchmarks
openff-bindingenergies
(General) – Fate of PLBenchmarks?
JW – We generally give the original writer the choice to continue as main maintainer, or turn over “primary maintainership” to infrastructure team. In the latter scenario, new feature development won’t really happen. But in the former, some input will be required from the original scientist.
Got some very strange results in FE calcs last week. I tracked it down to charges. Turns out that I’d opened the SDF with maestro and done some coordinate changes, and maestro changed the atom ordering, but not the partial charge ordering. There doesn’t seem to be a way to prevent this reordering/clear the SDF key/value pairs when I add/remote atoms.
PB – Looking at ways to filter conformations with strong sterics. Currently looking at relative nonbonded energies for confs of the same molecule.
JW – Will need to zero charges. I can walk you through this after sprint planning today
Add Comment