[gmx-users] Re: question about free energy. Gromacs user

[David Mobley]

Hi,

> I have been reading your work about free energy calculations and it is
> impressive. I want to estimate binding free energy for a ligand/receptor
> complex and I would appreciate your valuable input on how to approach the
> system I am using.  It is the first time I attempt free energy
> calculations.

Thanks. I am also CCing the users list on this so (a) I don't have to
answer the same questions again for someone else (hopefully) and (b)
you might benefit from the suggestions of others.

> I am trying to estimate the binding free energy between a beta-sheet
> receptor and 2 drug molecules that bind to the receptor in a beta sheet
> layer fashion. Also an estimate of the binding of the receptor to itself.
> From reading in Gromacs manual and tutorials, I decided to do a potential
> of mean force calculation where I have defined 2 distances between the
> receptor and the ligand to perturbe during free energy calculations.

OK. I haven't actually done any of these PMF binding free energy
calculations yet, so I may not be the best person to ask (I assume
you've read Benoit Roux's paper on the subject? I believe it was in
PNAS around 2005).

> Here is the part i added in the topology to perturbe in free energy
> calculation using lamda:
>
> [ constraints ]
> ; ai   aj  funct  length_A  length_B
>   44   258   2     0.259    2.406
>   123  179   2     0.329    2.334
>
> The distances i choose to constrain are bewteen the ends of the
> receptor-ligand interactions.

Maybe it should be obvious, but (a) why are you constraining two
distances, and (b) are you sure your constraints aren't going to muck
with the internal degrees of freedom for the ligand? I would think one
would like to pull the ligand out of the receptor along some
particular direction, but without doing anything to alter its internal
degrees of freedom, or there would be a free energy associated with
changing its internal degrees of freedom that you wouldn't capture in
the PMF calculation.

> So my system will change from state A, where the ligand is bound, to state
> B where the ligand is unbound. And then doing discrete simulation of
> different lamdas to get the free energy.

If you want to use constraints to do this, you should use the pull
code. I have a colleague who was just trying to do PMFs in the way
you're describing and is just getting ready to submit a bug report
(something's messed up).

> As you can see I have choosen to separate the ligand from the receptor. I
> dont know if this will give me a good estimate of the free energy. would
> do u think I should consider? orientational restrains? plz advise.

In principle you can get a correct binding free energy with something
like what you're describing, assuming you can converge the
calculations.

> will the method of orientational restraints and thermodynamic cycle is
> better to use for my system. I understand this is up to me to decide what
> method to choose, But I am asking since i am not an expert and your
> input will be valuable.

Like I said, I've never used the method you're describing. In my view
the big pitfall is that you have to pick a reaction path, but that's
not hard for one or several systems. That is, assuming you can pull
along some specified direction relative to the protein. If not, you
would likely encounter big convergence problems as your ligand would
have to sample a spherical shell at that distance from the protein in
order to get converged results.

The two methods (my usual approach, versus PMFs) probably have
different pitfalls. I would probably prefer the PMF approach for
exposed binding sites with charged ligands, while my approach would
almost certainly be preferable for very buried binding sites, or if
you don't want to define a reaction path. It also makes it easier to
consider multiple potential bound orientations.

Best wishes,
David


> I will include a pdb file of one of the ligand-receptor complex if you
> want to look at it.
>
> Thank you in advance,
>
> Belquis Mothana
>