Re: accelerated MD as the engine for loop refirement

From: Chris Ing (ing.chris_at_gmail.com)
Date: Mon Jul 14 2014 - 16:53:07 CDT

I can comment on the "old-fashioned"-"no accelerated
sampling"-"unrestrained"-"CPU burner" approach.

I'm currently working a project studying loop conformations for a membrane
protein where a crystal structure with resolved loops already exists.
Previous studies have suggested that this protein's loops have a functional
role, so we took the modelling quite seriously. I built 1000 homology
models using MODELLER that attempted to reproduce NMR predicted dihedral
angles in those loops (with additional restraints we input into the
modelling energy function). I then did K-means clustering on the subset of
"best" homology models to extract 5 representative structures. I membrane
embedded those structures and ran all-atom explicit water simulations of
all models and the crystal structure (500ns * 2 repeats for each). I found
that after 500ns none of the loops of the homology models crystal structure
converged to a common conformation.

The take home message is that some loops are floppy and could potentially
sample a crazy amount of states depending on the initial conditions. The
best scenario is one in which you can argue that the conformation of loops
aren't important for your study so you can just build a single homology
model and call it a day ;)

-Chris Ing
University of Toronto

On Mon, Jul 14, 2014 at 6:16 AM, James Starlight <jmsstarlight_at_gmail.com>
wrote:

> Hi Tristan,
>
> most people suggest me to use REMD instead of aMD for loop refirement
> because AMD might seriously alter thermodynamics of the system (during
> rewealling for instance). This will result in the non-physical
> conformations of the refined loops because of the non-bolzman distribution
> pattern obtained after AMD. What do you think about it? Also It'be very
> interested to know Jeff's opinions about such topic because of its great
> experiuence in this method.
>
>
> James
>
>
> 2014-07-14 9:40 GMT+04:00 Tristan Croll <tristan.croll_at_qut.edu.au>:
>
> If you’re going to try an accelerated MD approach like this I would
>> recommend at the very least applying cispeptide restraints to all mobile
>> residues (with perhaps the exception of X-Pro).
>>
>>
>>
>> *From:* owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] *On
>> Behalf Of *James Starlight
>> *Sent:* Monday, 14 July 2014 5:05 AM
>> *To:* Namd Mailing List
>> *Subject:* Re: namd-l: accelerated MD as the engine for loop refirement
>>
>>
>>
>> oh sorry it was quite difficulte to send message from the mobile phone
>> :-) so I'd to specify my question.
>>
>>
>>
>> 1 )Regardless of the enhansed sampling method I wounder if the
>> simulation of membrane protein in principle will be possible placing the
>> protein within the water box (not into the membrane) and with the
>> application of the restraints on its all atoms (to prevent its unfolding
>> due to its contact with the water not with the membrane) apart from the
>> loops which should be refined (which are in both cases will be accesible to
>> water)?
>>
>>
>>
>> 2) Had someone been experienced with the rosetta package for loop
>> modelling in case of membrane proteins? On what general ideas such
>> refirement is based?
>>
>>
>>
>> James
>>
>>
>>
>>
>>
>>
>>
>> 2014-07-12 18:28 GMT+04:00 James Starlight <jmsstarlight_at_gmail.com>:
>>
>> Thanks alot!
>>
>> Some additional question:
>>
>>
>>
>> Assuming that I need to refine loops quickly of my membrane receptor what
>> should be expected if I 1) place the whole protein into the water box (no
>> membrane) 2) apply position restraints onto the all parts of this protein
>> which in fact must be embedded into the membrane to artifacts of its
>> contact with water (not membrane) in my model 3) run some enhansing
>> sampling engine to sample loops only in water box keeping all other parts
>> frozen. Could some artifacts in loops be arrise due to such coarse-graining
>> of the environment?
>>
>>
>>
>> James
>>
>>
>>
>> 2014-07-11 2:59 GMT+04:00 Gianluca Interlandi <gianluca_at_u.washington.edu
>> >:
>>
>>
>>
>> Why not just running accelerated MD and explicit solvent? That might
>> already confer an idea of the flexibility of the loops. Another alternative
>> is to use metadynamics (implemented as COLVARS in NAMD or through the
>> plumed plugin) and explicit water. Metadynamics should be more efficient
>> than accelerated MD once you have identified the correct collect variables.
>>
>> You can also try out implicit solvent on its own or couple it with REM.
>> But I would always test those things independently before combining them--001a1132edec33f67404fe2e5224--

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:36 CST