[gmx-users] how to deal with LINCS warnings despite prolonged stabilization of the system?

[Mark Abraham]

On Wed, Jul 3, 2013 at 1:23 PM, Anna Marabotti <amarabotti at unisa.it> wrote:
> Dear Mark,
> thank you very much for your answer. I'll try to do what you suggest, but
> what makes me crazy is that when the simulation stops, I have no indication
> at all about what happens to the system (or at least, I don't know how to
> search for these indications). I'd like to understand what happens to my
> system, to avoid groping like a blind.
>
> Following suggestions in
> http://www.gromacs.org/Documentation/Terminology/Blowing_Up, I analysed the
> following energetic terms that in my opinion should give me an indication of
> the variation of the intramolecular energy: Angle - Proper Dih - Improper
> Dih - LJ14 - Coulomb 14 - LJ (SR) - LJ (LR) - Disper.corr. - Coulomb (SR) -
> Coul. recip.- Constr. rmsd - Total - Potential - Kinetic - Temperature -
> Pressure - all the Coul, LJ 14, LJ-LR, LJ-SR: Prot-PRot, Prot-nonProt,
> nonProt-nonProt. For NONE of these energetic terms I found something
> anomalous (for example, the "spikes"), and the order of magnitude of these
> energetic values are comparable with the ones of the system that
> successfully completed the simulations. Did I monitor the correct
> parameters, or not? What should I expect to see for such problems?

Those sort of things are good for diagnosing an inappropriate
parameter set, e.g. forgot to scale a distance by 10 when converting
units. The more complex the system, the harder it is to be sure things
are done right, because the signal has lots of components. Trying
things out in vacuum phase can help, too.

> Another question: if the system is "blowing up" should I see an
> "explosion-like" behaviour visualizing the trajectories, or not? At present,
> the only thing I see is the jump over periodic boundaries, but I don't think
> it's exactly the same.

Jumps across periodic boundaries are just normal and expected. There's
no magic box. You can post-process the trajectory to make things look
good if/when you care.

Sometimes it is clear where something (parameters, configuration) is
not working because you can see a local hotspot of fast-moving atoms.
For example, a water molecule stuck inside an unsuitable pore when the
protein atoms are restrained will have no option but to sit there at
high energy. But once it can shove things around, it will, and now
maybe there's too much local KE for the discrete integration to cope
with. That particular case is usually solved by deleting the water,
but if it's a case of badly-packed side chains, then sometimes there's
no choice but to choose a tiny integration step and let physics relax
itself.

The system can't do a Hollywood explosion, because of the periodicity
and being in the condensed phase. If you want to see the "best
effort," I guess you could multiply the time step by 100, ignore
grompp pleading with your better sense, and see what happens :-) I've
never tried it!

> Final question: for your experience, can a bad parameterization of the
> protein ligand influence not only the ligand itself but also the protein
> determining the starting of LINCS WARNINGS on atoms that apparently are
> (relatively) far from the ligand?

That seems unlikely, but strange resonance things do happen (e.g.
winds rippling whole bridges). Taking the ligand bound to some minimal
component, solvating that and running in isolation would be an obvious
part of testing that the ligand parameters have some chance of being
suitable.

Seeing the whole backbone get unhappy at the same time would be a clue
that the position restraints weren't appropriate, or the time step is
too large.

Mark