[gmx-users] Ligand Internal energy

[Mark Abraham]

On 6/09/2012 12:08 PM, Tom Dupree wrote:
> I am trying to differentiate between several binding poses for a protein
> ligand complex.

Then you need to measure the free energy differences between them. No 
single configuration is ever going to tell the whole story, and maybe 
not even a part of it. One option might be to use thermodynamic 
integration to annihilate the ligands, so that you have a path between 
the states, but there's a big literature there that you'll need to look 
into yourself.

> Initially I tried the LIE method however its results do not followed the
> expected trend based on experimental data. I then looked at the raw
> interaction energies between the ligand and its environment (LIG-rest in
> g_energy). This gives better relationship but still fails to explain the
> experimental results. I now want to select some representative frames for
> analysis with external scoring functions.

Inasmuch as they may approximate the above, that's a good thought. But 
rather than try to guess what the "best" conformations are, run a 
simulation of the bound ligand, cluster the protein-ligand conformations 
(to approximate the free energy surface of the bound ligands), compute 
things based on the cluster centers, and reweight by the cluster 
populations. This is a crude approximation to the above.

> I would like to select the frames that have the strongest interactions and
> least ligand strain. How do I obtain the internal energies for the ligand
> alone? (bending, twisting, stretching etc?)

That ignores the strain of the protein... However, trjconv and tpbconv 
will allow you to carve matching subsets of your trajectory and .tpr for 
use with mdrun -rerun.

> Also should I include the
> LIG-LIG LJ and Coul energies?

Shrug, might as well add some more random numbers :-) Force fields were 
not parametrized to be additive in this way, so it's not a surprise that 
they mostly don't work in this way.

Mark