To provide more info:
I've just finished running 1ns of NVE and NVT MD with pmemd.cuda, and they
DON'T give the issue described, with CA RMSD < 1.8 in 1ns simulation.
The problems therefore appear to arise with a combination of pmemd.cuda,
NPT (ntb=2, ntp=1) and (my) CHAMBER prmtop.
I would prefer to run this system with NPT, as a conformational change may
occur. Nothing as large as unfolding though, just a small part of the
protein opening up. I'm using a fairly large water box around the protein,
so perhaps NVT would still be ok for this. Any comments appreciated!
--Marc
On 24 November 2011 10:10, Marc van der Kamp <marcvanderkamp.gmail.com>wrote:
> Hello,
>
> This is the main issue in one sentence:
> When I run NPT MD of a protein with pmemd.cuda and a CHAMBER prmtop, CA
> RMSD rises quickly and enormously (>10 Ang in ~100 ps), whereas this does
> NOT occur with pmemd.MPI.
>
> I am running a protein-ligand system using the CHARMM force field.
> I'm using AMBER11/AmberTools1.5, with the latest patches for pmemd.cuda
> (bugfix.17)
> I created the prmtop using CHAMBER from a CHARMM psf file.
> I have noticed before that when one is using multiple chains with residue
> numbers being reset to 1 in new chains, CHAMBER makes some 'mistakes' (see
> e.g. http://archive.ambermd.org/201104/0417.html), i.e.
> - the 1st water molecule is lumped in with the last solute
> residue/molecule (a ligand in this case), so the RESIDUE_LABEL and
> RESIDUE_POINTER sections miss out on this water and need to be adjusted.
> - ATOMS_PER_MOLECULE are incorrect because of the same reason.
>
> To avoid this behaviour, I used a simple workaround: I modified the
> CHARMM psf first to have all residues in the same segment ("ALL") and
> numbered them sequentially. (I still need to manually fix SOLVENT_POINTERS.)
>
> Thereafter, I started running simulations with the created CHARMM prmtop
> files.
> I first ran a short equilibration protocol on CPUs, essentially consisting
> of heating the system and doing some NPT equilibration with positional
> restraints on the CA atoms, and subsequently gradually releasing the
> restraints.
> Thereafter, I started running NPT langevin dynamics (ntb=2, ntt=3).
>
> Something I've noticed a little too late is that when using
> pmemd.cuda_SPDP (on a single GPU), the total and potential energies take a
> nosedive in the first few steps (something I haven't noticed before for
> other systems with exactly the same inputs but using AMBER prmtop files):
> Etot is ~1400 kcal/mol lower and EPtot ~1000 kcal/mol lower in just ~150
> steps (3ps). Furthermore (probably in response), the CA RMSD of the system
> rises really quickly, reaching 10 Ang after ~100 ps and 20 Ang within 400
> ps (this is a ~300 residue protein) - essentially the protein is
> 'denatured' in just a few 100 ps.
> As mentioned before, when I run the exact same input with the exact same
> system but with pmemd.MPI, total and potential energy stay near the value
> after equilibration and RMSD develops 'normally' (stays below 2 Ang in
> 1000ps)
>
> What could be the cause of this? Can it be due to a bug in pmemd.cuda when
> handling CHAMBER prmtops?
>
> Many thanks,
> Marc
> PS I'm currently running tests with pmemd.cuda_SPDP with modified input,
> essentially the equivalent NVT and NVE simulations.
>
>
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Nov 24 2011 - 04:30:04 PST
