Add cyclobutane parameters which produce puckered ring #65
Comments
|
Here's an update: I did a minimization with the off-ffcompare pipeline for cyclobutane.
I also tried running compare_molecule_energies.py with cyclobutane. When I used the smirnoff tailored to exactly replicate parm@frosst for AlkEthOH (Frosst_AlkEthOH_parmAtFrosst.ffxml) all of the energies agreed. When I used |
|
So, in other words, you've confirmed that there is a problem, @bannanc , and particularly that the problem might be with angles? (Also, you may only be seeing the FIRST different term because of when the error is raised; it's possible you want to try it with a flag set so it doesn't raise assertion errors so you can go ahead and see the energies/which other terms disagree. My recollection is that it will let you do that...) |
|
That cyan structure looks exactly planar. Could it be that there hasn't been anything to break the symmetry, and that it's just "stuck" in the planar energy maximum? If you feed in the MMFF minimized geometry, does it go back to planar? |
|
Ah, I see this was drawn from a simulation, so it probably already had its symmetry broken and went back to being planar during minimization. Sorry! |
|
@jchodera the @davidlmobley yes, basically I've confirmed the problem is what we thought it was. and in smirnoff there are only two. In GAFF there have a special carbon type for 4-membered rings (cy) and that has a different set of angles: It was easier for me to add a parameter to smirnoff and rerun in the pipeline I'm using then to find parm@frosst input files. If I add an angle parameter corresponding to CT-CT-CT, making the smirnoff parameters: I still get a flat center ring:
I'm working on a parm@frosst minimization right now to confirm, but I think this is an inherited issue. |
|
OK, so if this is an issue also present in parm@Frosst we would want to look at what GAFF/GAFF2 have done to get this right. |
|
Confirmed, parm@frosst also minimizes to a flat ring, I can try it with a substituent to see if that changes anything.
I'm pulling GAFF2 parameters since that is where we have gotten estimates for other additions to the force field. Since this is a completely new atomtype, there is a lot to parse, but as I see it these are the most general types we would want to add:
The full set of GAFF2 angles around cyclobutane rings is long so I'm not going to copy it in here right now. I'll try to get it sorted into the categories above and propose what we should add later this week. |
|
Oh, just as a note, there aren't many cy parameters for torsions, but they all seem to be the same as c3 (GAFF version of CT, tetrahedral/sp3 carbon). |
|
Here are my recommendations, the only changes are in angles, but I'll outline everything I looked at for posterity. Non-BondUnsurprisingly the LJ parameters for TorsionsThe only dihedral parameters GAFF2 has for BondAlmost all parameters here are the same as the corresponding values for One thing to note here is that the bond between two ring carbons has a slightly lower force constant. I am not convinced we need another parameter for this, at least a this point in time, but might be worth keeping in mind. AngleI started by organizing parameters from GAFF2 based on the categories below. For each I list the category, my proposed SMIRNOFF parameter and then show the relevant lines from internal ring angle external angle with only heavy atoms GAFF2 (1 ring atom): @cbayly13 this is a section I could use some help. In general I expect the external angles to be slightly larger than 109.5º. I defined our type based on that assumption, but there are some notable exceptions here, I've tried to define the atom type in those cases. Do you think it is necessary at this point to track down these exceptions and make another angle parameter for them? external angle with 1 heavy atom and 1 hydrogen GAFF2 (no ring - ring - hydrogen): external angle with 2 hydrogens |
|
TL;DR below this line in the frcmodish file for SMIRNOFF |
|
@bannanc - do you want to see what happens if you add those and run some dynamics, e.g. comparing geometries for these types of molecules resulting from new and old cases? |
|
I added the parameters and did a minimization (no dynamics).
|
|
🎉 🎉 |
|
Caught up on the issue; I think @bannanc 's reasoning and actions are sound. I am surprised our parm@Frosst cyclobutane is planar; we knew it should be puckered and I thought we put in the effort to make it so but I guess not. I am surprised it is modulated all by bond angles and not torsions but there it is. The GAFF2 parameters look overfit but divide into broad categories as @bannanc has done; with heteroatom substituents like N and S it looks like a couple of fairly general child parameters might be warranted. parm@Frosst was definitely wrong here; GAFF2 has led the way. Nice catch @davidlmobley and @bannanc . |
|
Thanks @cbayly13 |
davidlmobley
changed the title
|
Closed via openforcefield/openff-toolkit#83 and #66 |
@davidlmobley Noticed that when running simulations with smirnoff99Frosst on molecules with cyclobutyl groups the ring is flat, which we don't think is correct and isn't true in GAFF (or parm@frosst).
When we talked about this in person, David mentioned there are cyclobutyl groups in the AlkEthOH molecule set which means the carbons should be assigned CT. I'm going to start by double checking which parameters the molecules in AlkEthOH get with smirnoff99Frosst and which atom types those parameters are assigned. Then I'll look to see if the parameters agree.
I'm guessing that there are some kind of angle or torsion parameters with
@for ring bonds that might be messing the cyclobutane when they shouldn't.Mostly for my own reference, David has code for minimizing and running gas phase simulations with SMIRNOFF in the MobleyLab/SMIRNOFF_paper_code or we also have minimizations in the MobleyLab/off-ffcompare repo which is public.
The text was updated successfully, but these errors were encountered: