[gmx-users] Re: PME (long range interaction of sugar molecules)

[Erik Lindahl]

Paul Czodrowski wrote:
> Hi Erik,
> Hi Celine,
> 
> 
>>>Are long range interactions computed for each group separately
>>
> anywhere?
> 
> 
>>Hi,
>>
>>No - they are not calculated separately for PME, simply because it would
>>be extremely expensive to do it. We would have to create a separate grid
>>for each group, and then perform PME for all possible combinations of
>>grids. With 3 groups we would basically have to do PME 6 times each
>>step!
>>
>>Now, it MIGHT appear in a future version, but the smartest thing is
>>definitely to store your trajectory, rerun mdrun and calculate the
>>energies for the stored frames. When you do this you can simply exclude
>>all interactions except sugar-sugar.
> 
> 
> Can't yout do this via an index.ndx-file?
> In my case, I want to consider the binding affinity of a ligang with a
> protein. When I run grompp & mdrun I use the switch "-n index.ndx" (in
> which I defined the ligand and the protein in 2 different groups, which
> appear at the end of the index.ndx).
> When I run g_energy, I can calculate the interactions between the protein
> and the ligand. Am I doing something wrong?
> 

It doesn't partition the long-range (lattice) part of the energy if you 
use PME, for the reason I described above. The problem is that the PME 
algorithm works with ALL charges at once, not pairs of them.


> By the way: Is there a possibility to consider explicitlz hydrogen bonds?
> Is there any reference, where I can read more, how the interactions are
> being calculated?

Check the references of the forcefields; there used to be a couple of 
forcefields (early versions of Amber for instance) that had explicit 
hydrogen bonds, but all modern forcefields use the normal electrostatic 
interactions to account for hydrogen bonding too.

Cheers,

Erik