AMBER Archive (2009)

Subject: Re: [AMBER] Can't reproduce RESP charges

From: FyD (fyd_at_q4md-forcefieldtools.org)
Date: Tue Jan 13 2009 - 05:52:13 CST

Dear Sanghwa Han,

> I am trying to reproduce the RESP charges of Ser for the Amber99 force
> field.

This problem has been already discussed several times in the AMBER
mailing list: You will NOT be able to reproduce charge values from the
AMBER force fields.

Indeed, charge values fitted from MEP are affected by many parameters,
among them: molecular (mol.) orientation, mol. conformation & the
fitting procedure itself. The differences can be quite important. How
much ? impossible to say... For instance, two different mol.
orientations of a minimum optimized by QM are responsible of a maximum
charge uncertainty of ~ 0.07 e. Conformations and different fitting
approaches lead to even higher charge differences (see the tutorials @
http://q4md-forcefieldtools.org/Tutorial; more will come). This is a
real problem in particular for new users who want to learn. In this
case, new users clearly need reference values which are not available.
Besides, this problem might take a particular importance for the
definition of initial conditions for MD simulations. But this, this
another story.

We developed the RESP ESP charge Derive program to solve this problem.
Charges developed using R.E.D. are perfectly reproducible with an
error close to +/- 0.0001 e. This means that if you decide to use
R.E.D. your results will be reproducible by new users "that will come
after you".

We also developed the RESP ESP charge DDaataBase to store charge
values and force field libraries. The scientific community can
_freely_ use this database to share results and compare charge sets
and computational conditions. Thus, I can propose you to use data
available in R.E.DD.B. that you might try to reproduce & consider as a
reference. You could follow the strategy below:

- You use the AIB or TYR-OMe residues already available in R.E.DD.B.
as a reference. You try to reproduce the results obtained for those
residues to learn (modifying the charge equivalencing for chemically
equivalent atoms, the number of mol. conformations & the control of
mol. orientations).
- If you can reproduce what is already available, you won. You apply
the same strategy/a similar strategy/a more simple/a more complex
strategy (as you decide from what you learnt) for your new residues.
You submit your data in R.E.DD.B. for the new users "who will come
after you".
- You run MD simulations with your new FF libraries. If you can
explain what you want all is fine. This is done.
- If your MD simulations are not satisfying, you change your FF/charge
derivation procedure. You re-submit in R.E.DD.B., you re-run MD
simulations, and compare trajectories.
- In your corresponding manuscript you criticize data in R.E.DD.B.
that lead to not satisfying results, and you describe which charge set
is the best.

Does it make sense ?

Now about the questions in your email email:

All the choices you described below seems correct:
> REMARK INTRA-MCC 0.0 | 1 12 13 14 2 8 | R => CH3CO
> REMARK INTRA-MCC 0.0 | 6 11 7 16 17 18 | R => NHCH3
> REMARK INTRA-MCC -0.4157 | 3 | K => N
> REMARK INTRA-MCC 0.2719 | 9 | K => H
> REMARK INTRA-MCC 0.5973 | 5 | K => C
> REMARK INTRA-MCC -0.5679 | 10 | K => O
> REMARK REORIENT 2 5 10 | 10 5 2 | 8 10 6 | 6 10 8

> 1. Is there any protocol that one can use to reproduce the Amber99 charges
> with a reasonable accuracy, say +/-0.05? Or what was wrong in my
> calculations?

No. My feeling is what you are doing is correct. However, difficult to
conclude without looking at the P2N file(s) you used.

> 2. Is it OK to follow the procedure C to calculate RESP charges of
> phosphoSer although the charges of "unphosphorylated" Ser do not agree with
> the Amber99 FF? Can I use these values for phosphorylated Ser in MD
> simulations that use the Amber99 FF? Homeyer et al. reported Amber FF
> parameters for phosphoSer [J. Mol. Model 12, 281 (2006)] but I could not
> find any comment on this issue.

If you need phosphorylated Ser, and charges are already published you
need a good reason for not using them. More complex/general approach ?

I hope the community will use R.E.D./R.E.DD.B./R.E.D. Server so that
such a problem can be solved once for all. R.E.DD.B. includes now a
peer-review system to re-inforce the quality of the charge values
generated by the users.

I am aware I do not directly answer to your questions. Sorry for that.
If you wish you can send to the AMBER of q4md-forcefield tools mailing
lists your P2N files, and we can discuss about the choices you made.
The R.E.D.-III.1 tools distribution includes now a tutorial with
examples of P2N files that should lead to low RRMS values for the
fitting step.

Do not hesitate to ask if you have more questions. I hope this helps.

regards, Francois

> In Tabel II of Cieplak et al. [J. Comput. Chem. 16, 1357 (1995)], the
> authors listed the following MM minimized conformations for C5 (extended)
> and aR (alpha helix) they used in RESP calculations. The angles with *
> appear to be constrained during MM minimization although the authors did not
> say that explicitly.
>
> Phi Psi Chi-1 Chi2
> C5 197 174 47 299
> aR -60* -40* 187 43
>
> I have tried five different calculations to reproduce the charges. Both
> geometry optimization and charge fitting were performed at the level of
> HF/6-31G* with the help of R.E.D. software
> (http://q4md-forcefieldtools.org/RED/). Two conformations were used with
> RBRA (except B, which used QMRA) as described in the R.E.D. tutorial. The
> P2N file for R.E.D. looks like this.
>
> ..
>
> REMARK INTRA-MCC 0.0 | 1 12 13 14 2 8 | R => CH3CO
> REMARK INTRA-MCC 0.0 | 6 11 7 16 17 18 | R => NHCH3
> REMARK INTRA-MCC -0.4157 | 3 | K => N
> REMARK INTRA-MCC 0.2719 | 9 | K => H
> REMARK INTRA-MCC 0.5973 | 5 | K => C
> REMARK INTRA-MCC -0.5679 | 10 | K => O
> REMARK REORIENT 2 5 10 | 10 5 2 | 8 10 6 | 6 10 8 => C-C-O/O-C-C and
> O-O-N/N-O-O
>
> ..
>
> Here is the result for Ser. The values with * and ** deviate from those in
> Amber99 FF by more than 0.03 and 0.05, respectively.
>
>
> Atom
> Amber99
>
> A
> B
> C
> D
> E
>
> CA
> -0.0249
> -0.0277
> -0.0267
> 0.0169*
> 0.0122*
> 0.0053*
>
> HA
> 0.0843
> 0.1112
> 0.1151*
> 0.0918
> 0.0893
> 0.0874
>
> CB
> 0.2117
> 0.1585**
> 0.1327**
> 0.0676**
> 0.0861**
> 0.1512**
>
>
> HB1
> 0.0352
> 0.0427
> 0.0496
> 0.0674*
> 0.063*
> 0.0377
>
>
> OG
> -0.6546
> -0.6215*
> -0.6108*
> -0.6159*
> -0.6244*
> -0.6074*
>
>
> HG
> 0.4275
> 0.4085
> 0.4048
> 0.4193
> 0.425
> 0.4025
>
>
> Amber99: Values listed in the Amber99 force field.
>
> A: Bond angles and bond lengths were taken from the equilibrium values
> listed in the Amber99 FF; All dihedral angles were fixed at the above values
> for C5 and aR. No optimization performed.
>
> B: Same as A but QMRA of Gaussian03 was used instead of RBRA of R.E.D.
>
> C: The starting structures were the same as A but geometry optimization was
> performed with Phi and Psi constrained only for aR. Chi's were allowed to
> vary.
>
> D: Same as C but Phi and Psi of both aR and C5 constrained. Chi's were
> allowed to vary.
>
> E: Same as C but all Phi, Psi, and Chi's were constrained for both C5 and
> aR. In this case only the bond lengths and bond angles were allowed to vary.
>
>
> Here are the questions.
>
> 1. Is there any protocol that one can use to reproduce the Amber99 charges
> with a reasonable accuracy, say +/-0.05? Or what was wrong in my
> calculations?
>
> 2. Is it OK to follow the procedure C to calculate RESP charges of
> phosphoSer although the charges of "unphosphorylated" Ser do not agree with
> the Amber99 FF? Can I use these values for phosphorylated Ser in MD
> simulations that use the Amber99 FF? Homeyer et al. reported Amber FF
> parameters for phosphoSer [J. Mol. Model 12, 281 (2006)] but I could not
> find any comment on this issue.
>
> Thanks, Sanghwa Han

_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber