[gmx-users] bad contacts on lipid tails

[Justin Lemkul]

On 6/14/17 1:05 PM, Mohsen Ramezanpour wrote:
> Dear Gromacs users,
> 
> I am doing simulations in HII phase in absence of any sugar or alkane to
> fill the gaps (in other words, HII in water)
> In few of my systems (which I made), there are lipids which their tails
> have bad contacts. Based on visualization, the CH2-CH2 bond from one tail
> in lipid 1 cross the CH2-CH2 bond in the tail from an adjacent lipid. They
> kinda stuck there.
> 
> Interestingly, these systems are running properly during EM step, as well
> as MD with a time step of 1 fs but fail with 2 fs, even after 45 ns of
> simulation with 1 fs. The lipid-lipid contacts cause LINCS errors for
> angles on those bonds and involved hydrogens, and eventually, the system
> will blow up.
> 
> This is what I understand after many tests to figure out the problem for
> this failure (segmentation fault (core dumped)). The topology of the
> molecule is correct and there is not water molecule which causes the
> problem. I also did simulations on one lipid both in the vacuum and in
> water. Everything is fine. The system setup is the only thing which I am
> suspicious about.
> 
> Is there any trick to:
> 
> 1) recognize these bad contacts? (I deleted few lipids but still, there are
> other lipids which cause the problem)
> 

Nope, aside from your eyeballs and inspection of maximum forces following EM 
that might point to a problem.

> 2) remove these bad contacts? (applying stronger bond and angle force
> constant, changing the LINCS parameters, or something like these?)
> 

Changing the force field is unwise.  Changing LINCS parameters can make the 
constraint algorithm more permissive and can overcome problems, but that can 
mask problems.

If your system is crashing despite "successful" EM, that means you probably only 
found a metastable state.  One thing that we have recently found is that the 
default settings for the GROMACS steepest descent minimizer can lead to 
premature termination but a decrease of initial step size to e.g. 0.002 (instead 
of 0.01, which is the default) does a much better job, especially when we 
compare it to the CHARMM minimizer (which is pretty bulletproof and does a great 
job of resolving nasty clashes that GROMACS doesn't handle well with default 
settings).

-Justin

-- 
==================================================

Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul

==================================================