Updates | SB Nothing too interesting, made a package that does custom nonbonded forces. Some issues with parameterhandlers, dependencies, vsites.
MT Some interoperability work, some cleanups needed with buckingham potential work. Want to sync up with JH on CustomNonbondedForces Worked on supporting interpolated parameters in certain OFFSystem representations. Currently these map to PeriodicTorsionForces in the OMMSystem. In the OFFSystem, for each unique interpolation that happens, the source of the data (partial bond orders) is saved. JW – Where is the partial bond order saved? MT – The actual bond order itself isn’t saved, instead the coefficients for linear interpolation are saved. (General) – Is there some way to make a parameterized system smoothly handle initially being set with a bond order of 1.9, and then have it be manually change 2.1? Could it store all the kX_bondorder{1,2,3} values? SB – PB and I were talking about having different interpolation schemes between bond orders (both in terms of having more points like k1_bondorder1.6, and also in terms of nonlinear interpolation) MT – I don’t think I enforce that the x values JW – I can look into how hard it would be to allow nonintereger bondorder values.
Worked on a stopgap solution to biopolymer topology (just storing an MDTraj topology alongside an OFFTopology). This let us write GROMACS output files that have residues included and still grompps Porting over toolkit tests into system for validation Will be porting over ParmEd test suite to validate conversions SB – Re: CustomNonbondedForce – Current path forward is to use a plugin to the SMIRNOFF spec. Working on buckingham potential | Github link macro |
|---|
| link | https://github.com/openforcefield/smirnoff-plugins |
|---|
|
. We should keep in sync on these representations.MT – In the short run, it’s probably not a good idea for JH and DCo to use the OpenFF system, since they’re not blocked by the lack of any capabilities. SB – Agree. But let’s stay in sync about the design, so that we can ensure that the OFF System can offer the range of functionality that we’re looking for. MT will join SB+JW meeting on Thursday to talk about object design.
JW Some interoperability meetings with Chodera, Shirts, thompson. Got fragmenter working with the new OpenEye toolkits (working = not crashing). I don’t think the numbers it procudes are “scientifically right” because it doesn’t do ELF10. Undecided on whether to release these changes. SB – I don’t think the new changes should be released. Probably needs a rewrite to be based off OFF Toolkit. JW – Unlikely that we can get this done from the infrastructure side in the enxt 2 months. The big lift would be in testing the changes. PB – I could look into this. I’m familiar with Josh’s large set of fragmenter results. JW – I could start this rewrite if PB can get it tested and validated. MT – This seems like infrastructure work, not science work. (Some discussion about how to prioritize work/how to ensure accountability) SB will attempt a rewrite of the useful subset of Fragmenter w/ toolkit backend, then PB will take over after a week.
Seem to be close to the finish line for contracting Andrew Dalke – Reviewed and negotiated contract. Benchmarking meeting on Wednesday. Industry partners are happy with progress. They have a large list of additional requests after the initial paper – We need to decide how to prioritize that and how to manage personnel later in the year.
PB Mostly WBO work. It turns out that, even when there are close matches with QM energies, the RMSD is high. This seems to overlap with JH’s work on whether conformational restraints improve fit. Currently optimizing just the new torsion parameters, not angles or bonds. Is this appropriate, or might this be a reason I see the geometry worsen? SB – I think all of these things are correlated to some degree. If we keep angles and bonds fixes,t eh torsions may get trained to compensate for a deficiency there. But this seems unlikely – JH is starting to get good results with that and careful selection of fitting parameters PB – The sorts of differences I’m seeing are RMSD of 0.2A, whereas the old FF get 0.1A.
Submitted genentech td 3 dataset to QCA dataset submission. This is blocked by the new psi4 release.
LW Mostly working on a manuscript w/ ANU Moved Parsnip libs from C++ to Python. Most of the pain was converting RDKit API. Some difference in availability of methods/exposed arguments, where the Pyhton API is a subset of C++ API. This work is in a local copy, I’ll upload it to GH soon.
|