"MULTISCALE manual; anisotropic scaling + solvent"

Poul Nissen nissen at KEMI.AAU.DK
Fri Mar 7 10:13:19 EST 1997


I have had a very similar case refining a protein-RNA complex at 40-2.9
A
resolution with X-PLOR. Too, I had great advantage of applying both
anisotropic
scaling and a bulk-solvent correction. Further, I also got a steady
increase in
the overall B upon successive rounds of these corrections after
refinement
cycles. What you may probably see, is that the Rfree against your native
data
(no anisotropic scaling, no bulk solvent corr.) has increased.
Erroneously, the
bulk solvent model is modelled to account for more and more of the
scattering, while
the model with increased B-factors accounts for less and less.
I came over this problem in the following way:
The model coordinates + B were subjected to overall B-factor (all atoms
grouped together)
and grouped B-factor refinement (mainchain, sidechain, phospho-ribose,
base)
against the native data (9-2.9 A), followed by anisotropic scaling 
(9-2.9 A 2sigma cutoff in scaling, 40-2.9 A output) and bulk solvent
correction 
(40-2.9 A). Finally, the model coordinates and B-factors were subjected
to grouped 
B-factor refinement (40-2.9 A) against the new corrected dataset (ani. +
bulk).
This corrected dataset and latest B-factor refined model was used for
map calculations 
and the proceeding refinement job was performed with those corrected
data also.
In this way, the successive increase of the overall B is avoided, and
one can on each modelbuilding/refinement cycle check Rfree against the
native data, independent
of corrections based on the model itself (almost, at least....). 
In my case, Rfree dropped by typically 5% upon anisotropic scaling and
bulk solvent
correction. However, I was most concerned with the Rfree calculated
against the
native data. This was my true measure of phase-improvement.
Note, that the "baoverall.inp" should be slightly changed, so that you
use the higher resolution data for the anisotropic scaling, while
applying
this scaling to the low resolution data also (use two d_min cutoff
variables, like
d_min_scale and d_min_out).
I have tested different scenarios of this cycle and have found that the
above
order of corrections is the best (e.g. bulk solvent corr. followed by
anisotropic
scaling leads to miscorrections). Further, I have tested the effect of
using
different sigma cutoffs in the anisotropic scaling and have found that a
cutoff of
2 sigma is optimal.

Poul Nissen
Laboratory of Macromolecular Crystallography
Institute of Molecular and Structural Biology
Aarhus University
Denmark



More information about the X-plor mailing list