Hi Andrew.
I was wondering what scripts are typically used when you are
close to
finishing the structure calculation. I am currently using
refine.py and it
outputs an ave.pdb file. I also have found a non-python script
average.inpin analyze_sry folder in eginput of the xplor-nih main
folder. I am looking for a protocol or method for determining the
final set of
structures when you are very close.
The script refine.py is pretty much what you want, I think. The
protocol should give you converged structures with good (maximal)
satisfaction of the experimental data. A lot of people would follow
the initial fold-determination with refinement in solvent nowadays,
but you don't need to.
Also, how many structures are typically done in the final structure
calculation? 100? 200? Of that what percent is acceptable to
submit as a
final cluster 20%, 10%?
This is an open question as far as I know. There are some (NMR-
determined) PDB entries that are single-conformer! The idea of an
"NMR ensemble" is mostly to illustrate the self-consistency
(precision) of the calculation. The most interesting issue is the
accuracy of the ensemble, however, and there is no general way to
assess structural accuracy. That's why the level of compliance with
target covalent geometries (like Ramachandran stats) are almost
always reported (although this practice is not satisfactory in my
opinion).
To "really determine" an NMR structure, the initial coordinates
should be an extended conformation, or some other unfolded
conformation (say, uncommenting the protocol.genExtendedStructure()
line in refine.py). 20% convergence in a calculation like that would
be pretty good, I think, so if 20 of 100 structures were low energy
and self-consistent, you could publish them. If you're getting a much
smaller number of converged structures, you could tune the
calculation protocol to increase the amount of convergence, anyways.
Making the hot phase longer, or hotter, or making the cooling phase
longer, or slower, often changes the observed level of convergence.
The arguments to simulationTools.StructureLoop() like
"averageTopFraction" or "averageTopNum" allow you to adjust how many
of the structures are considered acceptable.
Simulated annealing is a method for minimizing a multidimensional
target function, not a model of a real physical process, so there is
no general relationship between NOEs and the precision of an
ensemble. This is true even under a valid assumption of a single-
conformer interpretation of the NOEs. This concept is elegantly
demonstrated in "Conformational variability of solution nuclear
magnetic resonance structures." J. Mol. Biol. 250, 80-93 (1995). I
think the "information content" of a given fold's topology is a major
piece of the puzzle, as investigated in "Quantitative evaluation of
experimental NMR restraints." J. Am. Chem. Soc. 125, 12026-12034
(2003); you may want to try the QUEEN software (http://
www.cmbi.kun.nl/software/queen/) to feel/look more convinced of your
structure's self-consistency.
I hope that helps a bit,
Ryan
Ryan M.B. Hoffman
PhD Candidate (lab of Brian D. Sykes)
Department of Biochemistry,
University of Alberta.
_______________________________________________
Xplor-nih mailing list
[email protected]
http://dcb.cit.nih.gov/mailman/listinfo/xplor-nih
