Page 178 - Macromolecular Crystallography

MODEL BUILDING, REFINEMENT, AND VALIDATION 167

is the ’native’ dataset. Phase extension procedures software exists or may exist that offers similar or
and cross-crystal averaging should allow to extend more advanced features.
phases to the highest available resolution in all but
some extreme cases. We start with emphasizing a Very high resolution (above 1.4 Å): Automated model
few important issues: building with ARP/wARP should work even
with relatively poor starting phases since the
• Checking the data quality is strongly recom- free-atoms concept is particularly well suited
mended: inspection of the Wilson plot and data for high resolution maps. Due to the large
reduction statistics is very useful in judging the number of diffraction data at this resolution,
extent of the resolution to which the data can realis- the observation-to-parameter ratio is sufficient
tically be used. Pathologically bad data, for example to allow detailed model parameterization and,
those from a split crystal, twinned data, systemati- together with stereo-chemical restraints, is high
cally incomplete data, low resolution overloads will enough to correctly drive the optimizers. Refine-
always make model building and refinement hard if ment should generally be carried out with
not impossible. anisotropic ADPs using REFMAC or SHELXL.
• If experimental phases are available, getting the The use of strong NCS restraints may only harm.
best out of them before embarking on model build- Double conformations of side chains and even
ing is worth the effort and will definitely pay off. double conformations of main chain should be
Using software like SHARP (de La Fortelle and detectable in the density and can be modelled.
Bricogne, 1997) which puts particular emphasis on Solvent networks (except overlapping networks
detail and proper statistical treatment can be highly with partial occupancy) can very well be modelled
recommended. automatically.
• Automated procedures typically underperform High resolution (between 1.4 and 2.0 Å): Automated
compared to a reasonably experienced crystallog- model building with ARP/wARP should work
rapher. If an automated package did not provide with most phase sets. RESOLVE, which uses a
a solution, thinking about the problem and adjust- template-based rather than atom-based approach,
ing the strategy to the particularities of the case may should also perform well but may be computa-
lead to a success. Even if an automated structure tionally more consuming. Refinement can best
solution package provided good initial phases but be carried out with REFMAC or PHENIX using
automated model building failed, careful inspec- isotropic ADPs since the amount of data is no
tion and improvement of the phasing step can longer sufficient for an anisotropic description of
help subsequent automated model building to work atomic displacement parameters. The use of TLS
better. (Winn et al., 2003) is highly recommended.Ause of
NCS restraints should be critically evaluated and
in most cases the refinement can proceed without
11.4.1 Model building and refinement starting them. Double conformations of side chains should
from experimental phases
be visible and modelled. Ordered solvent can be
In the following text we list very general case stud- modelled automatically.
ies for native data extending to different resolutions. Medium-to-high resolution (between 2.0 and 2.4 Å):
The recommendations apply to software versions Automated model building with ARP/wARP or
that were available in mid-2006; it is likely that rapid RESOLVE should work for good phase sets that
progress in software development will make auto- produce clear maps easily interpretable by a
mated model building considerably more powerful, human. Such maps typically have average FOMs
particularly at lower resolution. The suggestions after phasing well above 0.6, which after density
about refinement are sufficiently general but, again, modification may rise up to 0.9 or higher. Refine-
the features of software are likely to change in the ment should best be carried out with REFMAC or
near future. The software packages referred to are PHENIX using isotropic ADPs. The use of TLS
those which the authors have experience with; other can still be very advantageous. NCS restraints