Re: AMBER: cutoff & speed-up of GBSA

From: Carlos Simmerling <carlos.simmerling.gmail.com>
Date: Wed, 30 May 2007 17:59:41 -0400

GBSA is sander is not efficient for large molecules.
as has been stated in the past, calculations in explicit water
can actually be much faster since you are using a shorter cutoff and
because there is a nonbond pairlist.

try to use GB in pmemd, that may help with speed.

On 5/30/07, Wei Chen <cwbluesky.gmail.com> wrote:
>
> Hello, Amber users,
>
> I got a big protein that has about 25,000 atoms. I try to run an implicit
> MD using GBSA. Because the size of the protein, I have to use a small cutoff
> and rgbmax. In order to get an appropriate cutoff and rgbmax, I calculated a
> single point energy when varing cutoff and rgbmax, separately. In the
> attachment, the first two plots are electrostatic & Vdw energy verse cutoff
> when rgbmax was fixed at 16 Angstrom. The third plot is electrostatic energy
> verse rgbmax when cutoff was fixed at 16 Angstrom. The electrostatic energy
> is the sum of electrostatic energy in vaccum and polar solvation energy. In
> the first plot, I see the electrostatic energy converges when cutoff >= 14
> Angstrom. Actually, the electrostatic energy is - 59497.1116 kcal/mol
> without cutoff. So it is very close. In the second plot, Vdw energy has
> larger variation than elec. energy. I think the Vdw force should vary
> slightly considering it is close to isotropic. Am I right? According to
> first two plots, a cutoff >= 14 Angstrom should be enough for my simulation.
> In the third plot, the elec. energy varies a lot verse rgbmax. So the elect.
> energy is much more sensitive to rgbmax. It seems to me that a rgbmax >= 24
> Angstrom should be used ( the elec. energy is - 58980.5952 kcal/mol with
> rgbmax=999). Is this reasonable?
>
> Another problem is with the speed of GBSA. I set cutoff=16 and rgbmax=16.
> Control parameters as follows:
>
>
> ----------------------------------------------------------------------------
>
> cat <<eof > mdin
> heat up structure
> &cntrl
> imin=0,
> ntc=2, ntf=2,
> cut=16.0, ntb=0, ntr=1,
> nstlim=10000, dt=0.002, nrespa=2,
> ntt=3, gamma_ln=1.0, tempi=0.0,
> igb=2, saltcon= 0.15, gbsa=1,
> intdiel=1.0, extdiel=80.0, rgbmax=16.0,
> nscm=1000,
> ntx=1, irest=0, ntpr=1, ntwx=100, ntwr=100,
> ntwv=500, nmropt=1,
> /
>
> #
> #heat up algorithm:
> #
> #from steps 0 to 10000: heat the system from 0K to 300K
> #
> &wt type='TEMP0', istep1=0,istep2=10000,value1=0.,
> value2=300., /
>
> &wt type='END' /
>
> keep all atoms restrained
> 5.0
> RES 1 1642
> END
>
>
> --------------------------------------------------------------------------------------
>
> I used SHAKE and set nrespa=2 in order to speed up. But it was still very
> slow. It is about 35 s/step on a single 600MHz R14000 CPU. As a comparison,
> I ran an explicit MD in a water box that includes ~200,000 atoms, 8 times
> bigger. I used Amber force field, but ran it in NAMD with cutoff=12. This
> costs ~20 s/step on the same single CPU. I searched on the mail list.
> Somebody also found that GBSA was slow for a big molecule. I am wondering if
> there is any way to speed it up today.
>
> I am appreciate for any response.
>
> Wei Chen
>
>
>
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Sun Jun 03 2007 - 06:07:16 PDT
Custom Search