From: Giacomo Fiorin (giacomo.fiorin_at_gmail.com)
Date: Fri Jun 21 2013 - 09:59:06 CDT
Hello Francesco, there are three answers to this:
I. Pick a simple distance variable between the COM of the drug and the COM
of the binding site (or the entire drug if you don't know it) and adapt
Usage Case 6 by using separately, or together if applicable, metadynamics,
ABF, thermodynamics integration (TI), SMD + Jarzyinski, umbrella
sampling... If the distance doesn't work, examine the trajectory to see
what transformations occur in the protein, and look for collective
variables to replace the distance, or to be added to the algorithm.
Learning from the first trials is always the best lesson, as mentioned
also in the abstract of the paper you cite. Also consider the possibility
of using alchemical free energy perturbation (FEP): comparing apples from
different grocers is useful, but so is keeping in mind that there are
oranges around as well!
II. You mention a specific variant of metadynamics, but for every method
you should trace down its historic evolution over the years. For example,
there is a 2013 paper by the Parrinello group with a new method for drug
binding called "funnel" metadynamics, which by my inaccurate count is the
7-th iteration of Laio's and Parrinello's work:
1) step-wise metadynamics (Laio et al, 2002)
2) continuous metadynamics (Iannuzzi et al, 2003)
3) well-tempered metadynamics (Barducci et al, 2005)
4) temperature-exchange metadynamics (Bussi et al, 2006)
5) bias-exchange metadynamics (Piana and Laio, 2007)
6) reconnaissance metadynamics (Tribello et al, 2010)
7) funnel metadynamics (Limongelli et al, 2013)
For the last three in particular, your only option at this moment is
PLUMED. I must warn you that this list can very well be *incomplete*...
A) For example, I considered the paper you cite an application paper, but
there is undoubtedly a lot of useful methodological information. B) There
is also related work by other groups: see e.g. the mid-90s papers by
Grubmuller and van Gunsteren, which were focused on protein conformational
changes. C) I may very well have missed some papers by mistake (Davide:
can you complete the list if so? Thanks so much!).
III. Even when you know the binding site, getting an accurate free energy
difference is not trivial, and certainly consumes quite a bit of computer
time. For blind docking, you may need your own personal supercomputer and
even then, that may not be enough... I would definitely recommend that you
use a mixed approach with cheaper computational methods (i.e. docking) to
first screen out really unsuitable candidates for the binding site.
Giacomo
On Fri, Jun 21, 2013 at 4:50 AM, Francesco Pietra <chiendarret_at_gmail.com>wrote:
> Hello:
>
> May I ask which one of the examples provided with the recent "Using
> collective variables to drive molecular
> dynamics simulations" by Fiorin-Klein-Henin could help with problems of
> blind ligand-protein docking? I am comparing with "Sampling protein motion
> and solvent effect during ligand binding" by Parrinello and associates
> www.pnas.org/cgi/doi/10.1073/pnas.1112181108, which was based on
> well-tempered metadynamics. This means that I am considering whether such
> problems of interaction between proteins and ligands can be treated with
> NAMD without the PLUMED plugin. It would simplify the life of NAMD users.
>
> Thanks
>
> francesco pietra
>
>
>
>
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:20 CST