[gmx-developers] turning charges on/off

[Jason de Joannis]

Quoting David Mobley <dmobley at gmail.com>:

> Jason,
>
>> No. I am planning on doing mutations between molecular species A and B.
>> Our idea is to have duplicate molecules with one on and one off. I know
>> how to turn off the Lennard-Jones, but I was worried about the charges.
>
> I am not sure I am *completely* clear on what you are doing,
Its rather involved and the paper is in press. Basically it's hybrid
MC/MD in the semi-grand canonical ensemble, or in other words the 
"mutation" ensemble.
> but if
> you are just talking about turning molecules on and off, you can
> definitely do this with the FEP code.
Good! Why reinvent the wheel.
> The FEP code works fine with PME
> as long as both A and B states have the same net charge, which should
> always be the case for you unless the molecules you are
> inserting/deleting have different net charges.
No problem for me.
> (In some sense, it
> works 'fine' anyway, in that the algorithm works fine, but there would
> be a contribution to the free energy due to turning on the netralizing
> charge used by PME, which is probably not what you want).
Not an issue since I don't need the free energy.
>
> To give you an idea of a "standard" calculation of this sort, I
> commonly do hydration free energy calculations where I first turn off
> the charges on my molecule in water, then turn off the Lennard-Jones
> interactions between the molecule and water, and then turn the charges
> back on in vacuum to get a hydration free energy. This works fine with
> PME. It sounds somewhat similar to what you are thinking.
Sort of, yeah. But I will be going from A to B directly and in one 
step. In other words lambda going straight from 0 to 1.
>
> On an unrelated note, if you are thinking about making MC moves
> between different molecules, you may need to think carefully about
> whether the phase space overlap is good enough to get anything
> meaningful back. As you're undoubtedly aware, we typically do these
> calculations with MD by adding intermediate states (  i.e. states
> where the molecule is 1/3 A and 2/3 B, 1/2 A and 1/2 B, 2/3 A and 1/3
> B, and so on), as the overlap between the A and B states is generally
> poor (different solvent configurations, etc). This would probably show
> up in MC with poor acceptance ratios, I suppose.
I know what you are talking about. I have done this stepwise business 
with both MC and MD in the past. In MC free energy simulations you need
to worry about overlap in distribution of states (e.g. acceptance ratio 
method). But since I am not interested in the free energy or its 
derivative
this is a non-issue. My acceptance rate will be dictated by the efficacy
of the configurational bias algorithm.
>
> Best wishes,
> David Mobley
> UCSF
Thanks for the insight. So far it seems that the FEP code will suit my
purpose, but I will probably have to adapt it so that only certain 
tagged molecules will have a lambda=1 and the rest will have lambda=0.
That doesn't sound unreasonable eh?


-- 
\ Jason de Joannis  \
\ Emory University  \
  \ jdejoan at emory.edu \
   \ 404-402-1332      \