I agree, plain MD might not be the best way depending on what you're trying
to learn. Using NEB and/or umbrella sampling might characterize the
transition if you know what it should be. Also a weighted ensemble approach
might let you explore the path (see WESTPA for example). If you're just
wondering if plain MD is capable, you might try an implicit solvent model
with reduced viscosity (igb=8 for example).
On Mon, Jan 20, 2020 at 2:24 PM Gerardo Zerbetto De Palma <
g.zerbetto.gmail.com> wrote:
> Hi! Maybe you should try steered MD or de umbrella sampling, given that you
> have structures of the bound and unbound protein. Just a thought!
> Regards!
> Gera!
>
> El lun., 20 de ene. de 2020 16:16, Homeo Morphism <
> homeo.morphizm.gmail.com>
> escribió:
>
> > There's a certain protein. When a ligand binds to it, it experiences
> > massive confomational transition: its certain domain covers distance that
> > is on the order of several dozens of angstroms and for some individual
> > residues it's even more.
> >
> > PDB structures for both ligand-free and ligand-bound (that is after this
> > massive conformational transformation has occurred) proteins are
> available.
> > There's also a set of experimentally verified mutations that make this
> > protein constitutively active.
> >
> > I've been trying to model the conformational transition of this
> > constitutively active protein with molecular dynamics and I'm failing at
> > it...
> >
> > First, I tried to use so to say ordinary molecular dynamics - tip3p
> water,
> > NPT with Langevin thermostat, SHAKE enabled, 2 fs timesteps, etc. I got
> > nothing, but perhaps this is expected.
> >
> > I then moved to accelerated MD, heating and cooling, just heating and
> > keeping it at high temperature, etc. Nothing. My accelerated MD attempts
> > are especially worthy of being mentioned: I was slowly raising boost with
> > each restart and the boost got so high that secondary structures began to
> > denature but there was no sign of the domain that is supposed to move
> even
> > beginning to move.
> >
> > Has anyone tried something similar with AMBER?
> >
> > Is my failure simply due to my not having found the right parameters and
> > modes of the simulation? There's still a replica exchange method that I
> > haven't tried yet. Self-guided langevin, etc.
> >
> > Or are there some fundamental constraints that preclude these large
> > transitions from being modeled?
> >
> > Any recommendations?
> >
> > I use pmemd.cuda and it delivers 100-150 ns per day on our GPUs for the
> > protein in question, so it's not Anton 2, but we aren't slow either.
> >
> > Thanks.
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Jan 20 2020 - 12:00:01 PST
