[gmx-users] Energy minimization: problems with ramachanrad score

[Thomas Evangelidis]

I 'm trying to relax my homology models from steric clashes, and while
searching for the appropriate minimization scheme, I came across this old
thread:

http://lists.gromacs.org/pipermail/gmx-users/2007-April/027043.html

The authors in the cited paper have created near-native structures as a test
set with RMSD 1.06 ± 0.14 Å over the native ones. Then they ran energy
minimization in vacuo with l-bfgs for 10000 steps or until convergence to
machine precision and they found that ordinary MM potentials (AMBER99,
OPLS-AA, GROMOS96, and ENCAD) showed no significant improvement on the
structures. On the contrary the last 3 were found to move the conformation
away from the native state by 11%, 40%, and 44% respectively. They also
tested their own 3 hybrid Knowledge-Based / Molecular Mechanics (KB/MM)
potentials and found that the best performing was moving the structures by
11% closer to the native fold.

Essentially they claimed that energy minimization with the ordinary MM
potentials is rubbish! I have one remark to make on this. Their decoy set
comprised no-native protein conformers with RMSD ~1A. I'm wondering if the
results will be similar for RMSDs ~3A as in my case where I'm building
homology models with average sequence identity to templates 25-30%?

And with respect to the original question, if you overdo it with energy
minimization the Ramachandran scores do deteriorate. However if you stop at
an intermediate step you can get the same score, bond lengths and angles but
with less steric clashes. Is this improvement plasmatic as the authors
claim, namely you have moved away from the native fold although
stereochemically the model look good?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20100617/aafe7f09/attachment.html>