From: Gumbart, JC (gumbart_at_physics.gatech.edu)
Date: Wed Apr 06 2022 - 21:21:04 CDT

I’m not sure you need to be internally consistent with how the charges are derived. It makes sense for water-accessible ones to follow the CHARMM style if possible but then it’s probably much less important for inaccessible ones.

You say PMF of drug binding, but no experiment gives a PMF (well, except maybe AFM), so I assume you mean binding free energy, or maybe even just the Kd. I don’t think you can parameterize based on this alone though, but your idea to use conformational energies and IR spectra is good.

One last question: I noticed when optimising the bonded parameters that as the objective function decreases/converges, the distance of the equilibrium bond lengths/angles from the QM values increases! Does this behaviour make sense? I have tried weighting the geometry higher and tightening the deviation thresholds, but they seem to make little to no difference.

To be clear, do you mean the MM values in the minimized structure or the values in the parameter file? These are unlikely to be identical for anything other than the simplest molecules.

But anyway, don’t overthink it! Just do the best you can in the time you have.

Best,
JC

On Apr 4, 2022, at 3:04 AM, Daniel Fellner <dfel694_at_aucklanduni.ac.nz<mailto:dfel694_at_aucklanduni.ac.nz>> wrote:

I've found this to be the case with quite a few drug-like compounds, particularly ligands for the more lipophilic receptors (cannabinoid, steroid hormone, etc.). Just a few technical questions: Do all the heavy atom charges need to be optimised in order to ensure internal consistency? Or, is it okay to leave some charges which were derived by water-interactions (presumably) from CGenFF? Some of these charges have low penalties/low deviations from the QM data.

I did check the NPA charges using HF/6-31G* v.s. wB97X-D/6-311+G** and there is very little difference, so probably no need to change at this point (unless it's what's causing me issues in the bonded optimisation...)

In terms of the philosophy, the optimised parameters should reproduce accurately the PMF of the drug binding. It's a very lipophilic compound - mostly partitioned into membrane so water interactions aren't critical. They should also help distinguish the conformational energies for the purpose of med chem... I'll compare the conformational energies between QM and MM conformers to validate the latter, and we can get an IR spectrum to help validate the bonded parameters as well. When it comes to partial charges though, I'm not sure how I would go about validating those.

One last question: I noticed when optimising the bonded parameters that as the objective function decreases/converges, the distance of the equilibrium bond lengths/angles from the QM values increases! Does this behaviour make sense? I have tried weighting the geometry higher and tightening the deviation thresholds, but they seem to make little to no difference.

Thanks again,

Daniel Fellner BSc(Hons)
PhD Candidate
School of Chemical Sciences
University of Auckland
Ph +64211605326

On Mon, 4 Apr 2022 at 08:41, Gumbart, JC <gumbart_at_physics.gatech.edu<mailto:gumbart_at_physics.gatech.edu>> wrote:
There was literally no way to do it with the water-interaction approach for a number of atoms in that molecule, so we just had to do what we could. We did indeed fix some hydrogen charges according to standard CHARMM style.

With regards to the QM method, I don’t think you need to stick with HF, necessarily? Play around and see how sensitive the numbers are to this choice. You need to decide what you’re optimizing for, which will be the real test of whether it worked or not.

Best,
JC

On Apr 1, 2022, at 3:52 AM, Daniel Fellner <dfel694_at_aucklanduni.ac.nz<mailto:dfel694_at_aucklanduni.ac.nz>> wrote:

Oh great, thanks! I didn't realise RESP was suitable for deriving CHARMM-compatible partial charges. I did notice FFTK has a RESP option now but I thought it was just for AMBER.

I assume the procedure is to put static restraints on the aliphatic and aromatic hydrogens not adjacent to a heteroatom (as per CHARMM protocol) and to any atoms you don't need to optimise? I gave it a try just now, and it does a good job for most atoms, though I do notice some unphysical charges on the buried atoms. I guess I'll set these with NPA first.

Regarding the QM level of theory, should I stick to the default HF/6-31G* for both the RESP and NPA calculations? My molecule is neutral and contains only C, H, N, F, O atoms unlike the metal complexes in your paper.

Thanks again for your help,

Daniel Fellner BSc(Hons)
PhD Candidate
School of Chemical Sciences
University of Auckland
Ph +64211605326

On Fri, 1 Apr 2022 at 04:51, Gumbart, JC <gumbart_at_physics.gatech.edu<mailto:gumbart_at_physics.gatech.edu>> wrote:
With regards to your first question, yes, absolutely it helps. But if an atom is completely buried, it’s definitely less sensitive to the water interactions, but not completely immune. The danger though is that it ends up as a “dumping ground” for excess charge while the more accessible atoms get optimized. Note that RESP suffers from the same problem. We discussed such challenges in this paper: https://urldefense.com/v3/__https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5980237/__;!!DZ3fjg!s_cefcymt0K3zNZ0AfhwLxo_EK4jD5WbbpFPHPgJQqdhOSmnNO5dM1rkUiqVcrimeQ$

We ultimately settled on a combination of RESP and Natural Population Analysis, with the latter for the buried charges.

Best,
JC

On Mar 30, 2022, at 3:13 AM, Daniel Fellner <dfel694_at_aucklanduni.ac.nz<mailto:dfel694_at_aucklanduni.ac.nz>> wrote:

Just had a few questions about charge optimisation for a difficult substrate.

If the atom in question is entirely solvent-inaccessible but has hydrogens attached, does the inclusion of target data from those hydrogens aid in fitting the occluded atom? I've seen this done in the EtOH tutorial but have read conflicting information.

As for occluded atoms with no hydrogens attached, how can these be optimised? The highest penalties in my compound are on entirely occluded (SASA <0.020 nm^2/mol) atoms. Is there any sense including them in the charges to be optimised without any target data?

Daniel Fellner BSc(Hons)
PhD Candidate
School of Chemical Sciences
University of Auckland
Ph +64211605326