Re: Parameter Optimization Advice

From: Brian Radak (brian.radak_at_gmail.com)
Date: Fri Mar 01 2019 - 13:57:15 CST

For a simple neutral, rigid molecule, you can think of the partial charge
problem in terms of what the dipole of the molecule ought to be. The
classic case is, of course, water. The gas phase dipole of a water molecule
is ~1.8 D, but in liquid it is >2 D (possibly much greater). This makes
sense when considering induced dipoles, etc. The difference is even worse
for larger, flexible molecules, since then the relative populations of
conformations ought to be quite different (the ones with a larger dipole
are favored in solution and disfavored in vacuum).

If you have a water model with an "enhanced dipole" (e.g. TIP3P, SPC/E,
literally, this is what the E means) that moves into the gas phase, there
is clearly a spurious energy penalty from the change in dipole and the
energetics may be quite wrong. There's actually a long literature on this
subject going back to the original SPC/E paper and more recent work by Bill
Swope and later Dave Cerutti (esp. in the context of protein force fields).

Anyway, bottom line, be careful constructing a model to describe both
phases. There are a lot of subtleties.

On Fri, Mar 1, 2019 at 2:31 PM McGuire, Kelly <mcg05004_at_byui.edu> wrote:

> Thanks Brian, those response are really helpful. This is quite a complex
> process, and can greatly affect results. On the last question, QMMM
> simulations, it sounds like if a bond isn't going to form, then don't jump
> to QMMM, but instead parameterize the molecule for use with an MM
> forcefield (with FFTK perhaps ) and run MM simulations. Molecules without
> transition metals or highly charged atoms for example can be modeled very
> accurately with one of the MM forcefields.
>
>
> By semi-vacuum environment, I worry about any energy calculations where
> the ligand is in our ion channel, which has very few waters, and then taken
> out of the channel into bulk water. For example, adaptive biasing force
> seems like it could be skewed depending on if the parameters are optimized
> for a gas state or condensed state, but what happens when the molecule goes
> from one to the other?
>
>
> *Kelly L. McGuire*
>
> *PhD Candidate*
>
> *Biophysics*
>
> *Department of Physiology and Developmental Biology*
>
> *Brigham Young University*
>
> *LSB 3050*
>
> *Provo, UT 84602*
>
>
> ------------------------------
> *From:* Brian Radak <brian.radak_at_gmail.com>
> *Sent:* Friday, March 1, 2019 12:18 PM
> *To:* namd-l; McGuire, Kelly
> *Subject:* Re: namd-l: Parameter Optimization Advice
>
>
> On Fri, Mar 1, 2019 at 1:14 PM McGuire, Kelly <mcg05004_at_byui.edu> wrote:
>
> I have gone through and used the FFTK parameter optimization for new
> molecules to simulate in NAMD. I have also used Gaussian independently of
> FFTK to get some parameters. I am trying to get a feel for best parameter
> optimization practices. Here are the questions I have though:
>
> 1) FFTK is mostly used to optimize water-molecule interactions and
> parameters, correct? But, if the molecule is in a semi-vacuum environment,
> say an ion channel, that has anywhere from 20-100 water molecules, then the
> molecule is probably interacting more with the sidechains and a vacuum
> environment than a bulk water environment. Does the FFTK optimization
> process do anything for interactions other than water molecules?
>
> This is a good point, and might warrant tests with ligands other than
> water. You can always make additional adjustments in order to best
> reproduce the energies and geometries that you think are most important.
> I'm not 100% sure what you mean by vacuum environment, but it's very
> difficult/impossible to have fixed partial charges that work well in gas
> and condensed phase since the nature of polarization is so different.
>
>
> 2) Partial charges are fixed in MD simulations, so even if I used a
> program like Gaussian to get good partial charges and then put those in my
> parameter files, how much can I trust and energy calculation process such
> as adaptive biasing force free energy profiles? In other words, can my
> choice of partial charges greatly effect my MD simulations? And, what
> process do you recommend for optimizing parameters?
>
> Without any testing, it's arguable that you shouldn't trust any partial
> charges of any kind. Try a simple problem and see if the results are
> qualitatively or semi-quantitatively inline with experiments. The specific
> charges may or may not greatly affect the answer (that might be an
> interesting result in itself!) and balancing Coulombic against
> Lennard-Jones interactions is the one of the most difficult problems in
> force field design. The CHARMM and FFTK philosophy is generally to get
> geometries and interaction energies correct, whereas the AMBER philosophy
> is to get the electrostatic potential correct and then separately adjust
> Lennard-Jones.
>
>
> 3) If QM/MM can be done, is that always the better choice because the
> interactions will be more accurate for the QM region/region of interest?
> In the QM region, do partial charges change, or are they still fixed during
> the simulation?
>
> It's a huge mistake to assume that QM is always more accurate than MM.
> There are all kinds of QM models, and especially QM/MM models, that are
> qualitatively incorrect compared to experiment (especially semiempirical
> methods). The only time QM is (trivially) guaranteed to be better is when
> bonds are breaking/forming. The most common implementations of QM/MM
> (electrostatic embedding) do include polarization and charge transfer in
> response to the MM region, but this is one-sided and thus has its own
> potential flaws.
>
>
>
>
> *Kelly L. McGuire*
>
> *PhD Candidate*
>
> *Biophysics*
>
> *Department of Physiology and Developmental Biology*
>
> *Brigham Young University*
>
> *LSB 3050*
>
> *Provo, UT 84602*
>
>
>

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2019 - 23:20:34 CST