Re: Inspiration for most efficient implementation of coarse grained forcefield.

From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Tue Dec 16 2014 - 09:30:40 CST

I'd strongly suggest using the MD-Go or SMOG features. Setting the
occupancy field of the GoCoordinates pdb file to zero turns off the Go
potential for that atom, so you can limit it to only C-alpha atoms.

Jim

On Mon, 15 Dec 2014, Ajasja Ljubeti~M wrote:

> Dear all,
>
> I would like to run coarse grained simulations of protein folding similar
> to Chavez et. al. <http://pubs.acs.org/doi/abs/10.1021/ja049510+> (here is
> the supplement
> <http://pubs.acs.org/doi/suppl/10.1021/ja049510%2B/suppl_file/ja049510%2Bsi20040602_123057.pdf>).
> The model is meant to study entropic barriers to protein folding (and is
> capable of predicting folding rates if the experimental melting temperature
> is known),
>
> Basically this involves taking only the C-alpha atoms from a known
> reference structure and determining the equilibrium values for bonds,
> angles and dihedral angles from the reference. Then a vdw-like potential is
> applied, that is specific for different atom pairs. Atoms that are close
> together get an attractive+repulsive part, all other pairs get only the
> repulsive part.
> [image: Inline images 1]
>
> My first thought was to have a single atom type and account for all the
> different bonds and angles using NAMD's extraBonds
> <http://www.ks.uiuc.edu/Research/namd/2.10b2/ug/node27.html#SECTION00086400000000000000>
> feature.
> But I don't see a way to have per atom pair specific vdw forces. So my
> first question is: *is there a way to have per-atom pair specific
> attractive/repulsive nonbonded potential? *(without resorting to TCLforces).
>
> The alternative is to have a separate atom type for each C-alpha residue.
> The I could write all the different values for the angles, dihedrals etc...
> directly in the parameter file. And the per-atom-pair vdw forces could be
> inserted using the NBFIX
> <http://www.charmm.org/documentation/c32b2/parmfile.html> terms of the
> charmm forcefield.
> So my second question is: *Are there any obvious performance issues with
> having ~1000 different atoms types in the parameter file and ~10000
> different NBFIX terms?*
>
> And the third (bonus, but perhaps the most important) question is: *Am I
> going about this wrongly? *I just saw, that NAMD 2.10 has MD-Go and SMOG
> simulations (which seem very very very similar to what I would like to do,
> except that it probably uses all atoms). So should I be using another
> feature of NAMD or another software package that I might have missed for
> such "Go"-folding simulations?
>
> Thank you for your help & best regards,
> Ajasja Ljubetič
> lbf.ijs.si
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:23:07 CST