DM – Fig 6.2 from thesis – Is this WBO dependent? Could we check this?
JH – I’ll check into this
JH – Some systematic issue with fitting H bound to boron that caused LJ radius to get huge (like 2.9 A) and charge to get negative
DM – Refitting BCCs could maybe prevent this.
DC – These weren’t using BCCs - Chargers were derived from higher-level QM and atoms-in-molecules approach. (D-DECK? approach)
DM – There are still pathological cases from density fitting approaches. Hirschfeld? has known issues. Esteban vohringer-martinez has done some work here.
DM – 2 pharma people indicated that boron would be helpful. Nobody asked for silicon.
JW – In addition to deriving nonbonded parameters, this could lead to a multiplicative increase in number of torsions.
DM – Agree, though the initial parameters don’t have to be great. Because right now there are NO options.
JH – Especially if these are introduced first through bespokefit
DM – How should we handle the lack of AM1BCC support for this? Can we use SMARTS-based ChargeIncrementHandler?
SB – That should be possible. We can kick off some QM calculations to do this. I’m not optimistic about finding property data of silicon and boron in thermoml.
JW – We can easily avoid missing parameter errors up front by putting in all-zero wildcard chargeincrement at top of list.
DC – Where could we find more data on physical properties for these molecules? Pharma datasets?
DM – Bill Acree had done some dataset compilation work for physical properties.
Â
15 mins
Thoughts on using XTB fitting from bespoke-fit for free energy runs
@Joshua Horton
JH – QM torsiondrives for JACS set is taking a while. xtb calcs finished quickly. So I’ll start working from xtb results, and will also do QM-based fit once it’s done.
DC – I’m interested to see how xtb results in vacuum perform in condensed calcs.
DC – Remind me which sets we’re going to run calcs on?
JH – TIG2 (or whichever one JChodera ran before). Could run calcs for all of them
SB – For free energy calcs, I think we’d been talking to OE about running these calcs through Orion. DM, can you update?
DM – OE has had schedule conflicts for regular meetings where we’d ahve talked about this. But they’re generally excited about this. However they’re quite busy right now preparing for a major software release. So this conversation may restart once they’re done with that.
DC – Would this complicate things if we wanted to use these results for the bespokefit paper? Would they expect to be coauthors?
DM – I don’t imagine that they’d object to this.
DC – So, we should go ahead and start generating FF files?
DM – Yes
SB – I wonder if we could use lilac for this? I looked over the pmx workflow and I think we could run this without Vyutas. Might need to justify use of pmx vs. perses on MSK resources. pmx results seem to be more comparable to previous results/stable so that would probably be best for now.
DC – So, among the attendees of this meeting, we don’t have people with expertise in setting up FE calcs. So this is the unknown that we’re thinking about right now.
DM – I have a postdoc (MPitman) who has run pmx locally, so she could help. Some usage tricks like hard-wired file names.
SB – DHahn should also have expertise in this. So we should have in-house expertise.
DM – We could schedule a quick chat with DHahn and MPitman on setup issues.
SB – Let’s loop in DDotson and other software folks since they’ll be interfacing with this as well.
(General) – Some strategic decisions to make with how we use/fork/fix PMX, where we get compute resources, whether we look to another FE framework
DM – One nice thing about pmx is that there’s less scope for uncertainty/implementation chagnes. Once the gromacs topology files are written, they’re written, and the simulations will agree within statistical error.
Action items:
JH will make up force field files from XTB results.
We’ll coordinate a PMX usage meeting with JH, DH, DD, MPitman, JW, SB, DM.
Â
Â
Torsion multiplicity results
@David Mobley
DM – Finding some cases where torsion parameters are being applied to motifs with different topological multiplicities. Eg, in figure below, the left molecule has t59 applied to a ninefold torsion, but on the right it’s a fourfold torsion.
DC – Would this affect bespokefit? I don’t think it wouldsince we expand the torsion multiplicies.
JH – Right, this shouldn’t affect us as long as we keep using specific SMRIKS. If we starter letting our SMIRKS get too general then we could hit this problem.
DM – PB, could you contact Megan and check this against QM results? Basically I’m interested in seeing if these result in higher errors.
