RE: Strange glitches when performing energy minimization in MDFF simulations

From: Tristan Croll (tristan.croll_at_qut.edu.au)
Date: Mon Feb 10 2014 - 18:24:37 CST

I've gone back over the last dozen simulations that involved energy minimization, and I'm seeing similar single-point spikes (ranging from 10^7->10^9 kcal/mol), seemingly randomly distributed in the middle of the minimization process, in about half of them. I can correlate this back to structural deformations in three. In one clear case, I have an arginine residue that looks completely normal at step 1000. Then there's a roughly 10^9 kcal/mol spike in the logfile at step 1062, the arginine is entirely distorted with the water molecules pushed away by about 5A... then by step 5000 the structure is back to normal and MD simulations continue with no sign of trouble. There are no warnings, and no error messages. I've found similar events in two other trajectories, involving entirely unrelated and >50A distant residues. If I have to guess at what's going on, I would say that each of these events correlates with the system losing track of the position of an atom, with the distortions occurring as it snaps back into place.

This is driving me crazy.

From: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] On Behalf Of Tristan Croll
Sent: Monday, 10 February 2014 4:21 PM
To: namd-l_at_ks.uiuc.edu
Subject: namd-l: RE: Strange glitches when performing energy minimization in MDFF simulations

Just found this in one of the logfiles. Something's clearly glitching, but where? Is this more likely to be a hardware or a software fault?

ENERGY: 11731207 70944.7907 51205.7902 19764.7817 195.5461 -2487642.8961 216593.6539 0.0000 606.7625 0.0000
  -2128331.5712 0.0000 -2128331.5712 -2128331.5712 0.0000 -7479.5403 -7431.4393 5044132.7704 -7479.5403 -7431.4393
ENERGY: 11731208 72427.9957 51488.6168 19766.0213 214.8399 -2490093.8226 1691328329.2127 0.0000 606.7798 0.0000
  1688982739.6436 0.0000 1688982739.6436 1688982739.6436 0.0000 93196322.9979 93196626.2239 5044132.7704 93196322.9979 93196626.2239
ENERGY: 11731209 70806.8876 51169.4804 19765.2833 192.8327 -2487178.9184 216259.2942 0.0000 606.7590 0.0000
  -2128378.3811 0.0000 -2128378.3811 -2128378.3811 0.0000 -7485.5098 -7444.8419 5044132.7704 -7485.5098 -7444.8419

From: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] On Behalf Of Tristan Croll
Sent: Monday, 10 February 2014 3:50 PM
To: namd-l_at_ks.uiuc.edu
Subject: namd-l: Strange glitches when performing energy minimization in MDFF simulations

Hi,

I'm currently running simulations in which I have a protein complex fitted within a ~10A cryo-EM map, and have been running "pseudo-equilibrium" simulations (with the MDFF constant dialled all the way down to 0.005) to allow it to rattle itself down to an energy minimum with weak guidance from the map. I've found that under these conditions it seems quite safe to remove all secondary structure, cispeptide and chirality extrabonds without anything going awry - except in the specific case where I perform an energy minimization at the start of a continuation run. For some reason, during minimization small, seemingly random portions of the structure (e.g. within a sphere of ~10A diameter) will jump into crazy configurations (bond lengths and angles well away from normal) before settling back to a "standard" arrangement (except sometimes with trans to cis or chirality flips). Apart from the MDFF forces, which shouldn't be nearly strong enough to do this (as I said, everything is very stable in MD without restraints), the rest of the simulation settings are very standard. I'm at a loss to explain what is going on, and this has thrown into question what I thought was >50ns of production run. Can anyone help?

Thanks,

Tristan

NAMD version is 2.9-ibverbs
Using CHARMM-36 parameters
Explicit water, 0.15M sodium chloride
I've checked that the system is neutral

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:05 CST