AMBER: RE: ظ RE: Asking for help about heating

From: Ross Walker <ross.rosswalker.co.uk>
Date: Mon, 24 Sep 2007 08:52:41 -0700

Dear Wang,
 
As I explained before please send such questions to the amber mailing list
(amber.scripps.edu) - there are instructions for how to sign up on the Amber
website (http://amber.scripps.edu).
 
To answer some of your questions:
 
1) The system I simulated has 1200 residues, how long the heating time
should be suitable ? How can I control heating the system slowly in amber?
Does it mean I need to add some keywords in input file or "slow" heating
only means heating the system long enough?
 
This could really be considered the definition of a hard question. How long
is long enough is something I think people in this field have been debating
since it's inception. Ideally you want to heat your system on a time scale
that is compatible with the rate of heating in experiment but this is
obviously way beyond the reach of current simulations. The key is to do it
slowly enough that you don't perturb your system in what is an unphysical
way that results in it moving into a region of phase space that would
otherwise not normally be accessible to it. Typically the better your
starting structure is the faster you can get away with heating. Something in
the range of 50 to 250ps is normally reasonable for most systems.
 
To control heating in amber you can use weight restraints to linearly scale
temp0 as a function of the step number. See page 199 of the amber 9 manual
for an example that you should be able to adapt to do this.
 
2) Does it make any difference for different heating steps , such as 0.5fs
and 2 fs ?
 
The key point you are referring to here I assume is not "heating steps" but
the time step. These are two very different things. Read up on the concept
of integrators and numerical solutions of Newtons equations of motion and
you understand this better. The key with the choice of a timestep is that
the acceleration is deemed to be constant over the course of a timestep.
Hence if your timestep is too big you get instabilities. However, if you
timestep is too short you end up just wastign computer time. Typically when
using shake a timestep of 2fs is reasonable. However, it may be necessary to
reduce the time step to something smaller, such as 0.5fs if your system is
unstable. This often occurs if you have a poor starting structure, from
homology modelling or a low resolution crystal structure. In this case the
initial acceleration on starting MD can be high and cause the system to fall
apart. Sometimes it is possible to get around these initial instabilities
and allow the system to relax away from this high energy point by using a
shorter timestep. So during the heating phase you could use say 0.5fs
timestep and then increase this to 2fs at the end of heating. For most
systems however this is not normally required.
 
Good luck
Ross


  _____

From: Wang ling yun [mailto:wanglingyun.wipm.ac.cn]
Sent: Monday, September 24, 2007 01:52
To: Ross Walker
Subject: ظ RE: Asking for help about heating



Hi Professior :

I am sorry to interrupt you, but I only want to get my answer.

       The system I simulated has 1200 residues, how long the heating time
should be suitable ? How can I control heating the system slowly in amber?
Does it mean I need to add some keywords in input file or "slow" heating
only means heating the system long enough?

        Does it make any difference for different heating steps , such as 0.
5fs and 2 fs ?

        The system is heated for 5ps in step of 0.5fs , will it make
unaccepted results ?

Thank you for your help !

Lingyun Wang



-----ԭʼ-----
ˣ "Ross Walker" <ross.rosswalker.co.uk>
ѷ 2007-9-11 7:31:24
ռˣ "'Wang ling yun'" <wanglingyun.wipm.ac.cn>
ͣ amber.scripps.edu
⣺ RE: Asking for help


Hi Lingyun,
 
I am happy to answer your questions but in future can I request that you
post queries via the amber mailing list (amber.scripps.edu) - see
http://amber.scripps.edu/ for details on how to sign up. By posting queries
to this list you'll often get answers more quickly and more importantly the
questions and answers end up in an archive
(http://amber.ch.ic.ac.uk/archive/) where other users can search them.
 
The question regarding heating your system vs hot starting it at 300 K
really depends on how stable a system you have to begin with and this
depends on how good your initial structure looks. If you have a well
equilibrated solution phase structure then you can probably get away with
starting it at 300 K. However, if you have a crystal structure, that you
have just solvated, by hot starting at 300 K you may find that there are
stability problems due to bad steric clashes etc. Or if the simulation
doesn't show any initial problems you may have ended up giving the system an
unrealistic perturbation. The problems are a function of crystal structures
being low resolution and/or solid state structures and the fact that leap
has to add a number of solvent molecules that can clash with protein atoms
etc. Thus a way around this problem is to heat the system up slowly in what
is a more controlled fashion so that the system has a chance to relax away
from what is an initial high energy state.
 
Note, for small systems, like small peptides in water, and also for implicit
solvent systems it is possible to just start them at 300 K without any
problems but it is so easy to heat a system slowly that you might as well do
it to be on the safe side.
 
In answer to your question about restraining the protein during initial
heating in explicit solvent this is not normally necessary. It is included
in the tutorial because in some cases, especially with homology modelled
starting structures, or structures where missing residues have been added
etc, it is necessary to use restraints initially to obtain a stable
simulation. Thus this essentially represents a worst case scenario.
 
The question of how long a simulation is suitable is extremely difficult to
answer and is largely a function of what you are looking to observe.
Typically the larger a system the more sampling you will need to do. Rule of
thumb these days is that simulations should typically be on the nanosecond
timescale at least but certainly one could write several Ph.D.s on this
question. For the heating stage it just needs to be slow enough that your
system isn't badly perturbed by this. Something on the order of 100ps or so
is probably reasonable for most systems.
 
With regards to different simulations giving different results yes indeed
this is true. Simulations started from different initial structures, started
with different random seeds, or even just run on different numbers of
processors will follow different trajectories. But none is more right than
any of the others. The issue is one of ensemble averaged behaviour. I.e. if
you have run long enough sampling then all of the simulations should
converge to the same result. If they do not then you may be sampling
behviour that is a function of your starting conditions rather than a aspect
of the system you are simulating.
 
Good luck...
Ross

/\
\/
|\oss Walker

| HPC Consultant and Staff Scientist |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
| http://www.rosswalker.co.uk <http://www.rosswalker.co.uk/> | PGP Key
available on request |

Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.

 


  _____

From: Wang ling yun [mailto:wanglingyun.wipm.ac.cn]
Sent: Thursday, August 30, 2007 19:11
To: ross.rosswalker.co.uk
Subject: Asking for help



Hi Professor :

    I am a new amber user , may I ask you some questions ?

    In a md simulation , is nescessary heating the system ? May I run a
simulation at 300K directly ? What's the role of heating ?

    If I do a protein simulation in explict model, is it necessary to
constrain the protein or not ? How long the simulation is suitable ? And do
the different simulations make different results ?

 

Thank you for help !

Lingyun Wang




-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Mon Sep 24 2007 - 17:45:31 PDT
Custom Search