On Mon, Apr 20, 2015 at 2:20 PM, George Tzotzos <gtzotzos.me.com> wrote:
> Jason,
>
> As I’ve come across similar situations, I’d like to chip in.
>
> In situations like this, would it help to do (i) clustering of the
> trajectory, and (ii) running MMPBSY.py on each cluster?
>
Again, it depends on what you hope to learn, and how you "run MMPBSA.py on
each cluster". The points I made about over-sampling lesser-populated
clusters by, e.g., taking the same number of snapshots from each cluster
remains (depending, of course, on the clustering algorithm you used).
Random sampling (or periodic sampling of a random, stationary distribution)
should give you the correct ensemble statistics (to a confidence determined
by the size of the subsample). Trying to cleverly use the ensemble to
determine how to sample that ensemble is just asking to introduce bias into
your binding free energy results, in my opinion.
What I'm saying corresponds to an assumption that you are using MMPBSA.py
to compute a binding free energy for a noncovalent protein-ligand complex,
and that you only care about that binding free energy. If you are instead
interested in the relative stabilities between two conformations (and you
don't feel that they are properly converged so you can't simply use their
populations), then some of these clustering-based approaches may be
appropriate.
All the best,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Apr 20 2015 - 12:00:02 PDT
