Oops, it seems like the sulfoximine moiety (which we never considered when creating and testing the atom typing rules) triggers a previously unknown bug. The S atom type should be SG3O2 rather than SG3O3 . This will result in different parameters and charges being assigned by analogy (though the differences in parameter values may not be that big). More importantly, different existing parameters will be used. In particular, the SG3O3 OG2P1 bond is special; SG3O2 OG2P1 should probably work better for your molecule. Also, the CG331 CG321 SG3O2 OG2P1 dihedral is significantly different from CG331 CG321 SG3O3 OG2P1 . The other differences (the CG31 SG3O OG2P1 angle and CG31 SG3O bond) will probably have a smaller impact. I'm sorry about this; this is why it's beta. I'll have it fixed in the next release.EDIT: In the meanwhile, if you post the mol2 file of your molecule, I will run it through a developmental CGenFF version that contains a temporary workaround for this bug.
Now about the charges:
- The -0.73 on C19 is absurd; this is a clear case of overfitting. It is strongly recommended to stick with the default -0.27 .
- To improve the transferability of the charges, it would also be recommended to fit the charges under the constraint that C9 should be exactly 0.09 more positive than C13. See RESI MESN in the main CGenFF distribution for a good example. Additionally, both C9 and C13 seem way more negative than could possibly be physically justified.
- Similarly, S11 is way too positive (which goes hand in hand with the carbons being too negative).
- C4 also appears to be more negative than it should be.
I think the above points need to be taken care of. Specifically, you want to re-optimize the charges on the C-C-SO-C part while staying reasonably close to the corresponding charges on MESN. In fact, I think you should set C13, C9 and C19 to the exact
same values as in MESN, since you don't have target data for them (I assume this because typically, all the interactions with aliphatic hydrogens are repulsive on molecules that have a net negative charge). Similarly, C4 should be set to the value in RESI DMEP and MP_1 (-0.17). As a side note, you might want to get rid of the 3rd decimal in the charges while you're at it. It's not really significant, it looks sloppy, and I've heard that some old versions of some MD programs wouldn't read it correctly.
Note that the above changes - especially the change in atom types but also the big change in charges - may require tweaking the bonded parameters (that's why we do the charges first).
Finally, carefully consider on which conformation you want to optimize the charges and bonds and angles. What is the difference in energy between the gauche (±60°) and trans (180°) conformations of the O-S-N-P dihedral? If the trans conformation were to turn out to be much higher in energy, it will be scarcely populated, and it may be worth optimizing the force field on a gauche conformation instead (provided
that the the water interactions are not much more hindered in this gauche conformation). The same goes for the S-N-P-O3 dihedral; consider if it shouldn't be trans rather than gauche. And then there's also the N-P-O-C and N-S-C-C ... What we usually do is pick a reasonably low-energy structure in which the water interaction sites are mostly unhindered, which often (but not always) corresponds to an all-trans or extended conformation. In some extreme cases, we even calculate water interactions on two distinct conformations and make a compromise when optimizing the charges.