Re: [AMBER] Interpreting MM-(GB)PBSA results

From: Josh Berryman <>
Date: Mon, 28 Jan 2019 16:53:15 +0100

For my 2 cents I'd just like to mention that while PBSA is sometimes quoted
as a reference for training GBSA models, in fact GBSA methods are often
more stable and handier, as they have no grid, and have more free
parameters for tuning to commonly-encountered cases (like a druglike in a
protein binding pocket).

For flexible or loosely bound systems you will need to make some sort of
entropy measurement, you might want to think about umbrella sampling,
thermodynamic integration or steered molecular dynamics (all of which work
with explicit solvent (better) as well as implicit). Implemented methods
abound, however none are computationally cheap. If you have access to GPU
accelerators then just using TI or steered MD to pull the drug in and out
of the pocket a few times is quite a neat and not-so-slow way to measure
the work.


On Mon, 28 Jan 2019 at 13:54, David A Case <> wrote:

> On Sat, Jan 26, 2019, Sundar wrote:
> >
> >I am performing MMPBSA and MMGBSA calculations for a protein-ligand
> >complex. Protein is small, containing 50 residues. GBSA is giving a Delta
> >Total of -5.2 and PBSA is giving 0.34 kcal/mol. How to interpret the
> >difference between these numbers? Why's GBSA binding affinity is stronger
> >than PBSA?
> This is not an unusual difference. You might read various reviews on
> MM/PB(GB)SA end point methods. It is in general much harder to get
> useful quantitative results than many new users expect.
> >
> >How long should be the simulation to get reliable results? How do I
> measure
> >the convergence?
> >Do multiple replicates of the simulation help to improve the accuracy?
> >
> >What if the ligand leaves the binding site during the simulation? Can I
> >include those frames as well in these binding affinity calculations? If
> >not, can I use some restraints to keep the protein-ligand bound together?
> Above are good questions, but not ones that can be easily answered on a
> mailing list, since there probably are no general answers. Don't be
> afraid to experiment. If you ligand is not stable in the binding site,
> then the assumption that you can correctly sample configurations from
> the end-point "bound" state is probably not valid.
> ....dac
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Mon Jan 28 2019 - 08:00:03 PST
Custom Search