Re: Newbie's question

From: Axel Kohlmeyer (
Date: Tue Aug 07 2012 - 03:38:50 CDT

On Mon, Aug 6, 2012 at 7:41 PM, Theresa H <> wrote:
> Dear all

dear theresa,

> I am a protein crystallographer with interest in MDS. Since we don't have
> any computational biologist in the department, I want to ask a few basic
> questions.
> I have the structures of two membrane proteins (protein C and D) and one
> soluble protein (protein B). Protein A can interact with protein C or D
> through protein B. The protein D also has a domain similar to protein B.
> Therefore I am interested in knowing why the protein A can't interact
> directly with protein D.
> 1. Is NAMD the right tool to answer this question?

i think you are asking the wrong question here.
it is notso much a question whether NAMD is
the right tool, but whether one (or more) MD
simulations can give you the answer that you
are looking for.

..and that is not an easy to answer question.
you have to realize that MD simulations by itself
don't provide "yes or no" type of answers. and
particularly not without careful planning.

the way people can extract useful information
from MD is more that you formulate a hypothesis
and see if an MD simulation can support or
disprove it. this often requires solid knowledge
in statistical mechanics. while the mechanics
of performing an MD simulation are typically
very simple, the planning and analysis are not.

also you have to keep in mind that you have to
deal with a large number of approximations and
assumptions. classical MD uses a phenomenological,
but still rather crude and inflexible model for interactions.
you typically can only afford to run simulations on
with a restricted number of particles and then often
are still limited to a nano- to microsecond timescale.
and often you would need multiple independent
trajectories and/or other enhanced sampling methods
to further reduce the statistical (not systematical) errors.

> 2. If so, do I need anything else apart from the crystal structures?

crystal structures are a beginning. often they are missing
pieces that need to be reconstructed and you also may
run into non-standard residues that will have to be parameterized.
since you have membrane proteins, you have the additional
complication of having to merge your protein model(s) with
a suitable membrane and equilibrate that (and also model
the potential drop across that membrane?). this overall can
be a somewhat straightforward process that is covered more
or less by the available documentation and tutorials, it can
also be very tedious and difficult, if you have problematic
compounds and need to do extra work.

> 3. How much computational resource is needed?

very likely: a *lot*. my recommendation, try to find
a collaborator with experience in doing such simulations.
there is nothing more "dangerous" in the simulation
business than doing simulations without the sufficient
guidance. the help and support that a mailing list
can provide is insufficient in that respect. a lot of
things require a significant amount of experience
and as a "lone ranger" you are likely to make all
the errors that everybody does initially, but are
liable to not notice them, which can potentially
result in a lot of wasted time.

in short, MD isn't as simple as it looks
and people try to make it look. it can be
a very versatile and powerful tool, but to
become that, it requires expertise that
is not always easy to get. mind you that
the system you are looking at and the
question you want to be answered is a
somewhat complex one.


> Thank you.

Dr. Axel Kohlmeyer
International Centre for Theoretical Physics, Trieste. Italy.

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:54 CST