Page 179 - Macromolecular Crystallography

168 MACROMOLECULAR CRYS TALLOGRAPHY

in this resolution range are still not absolutely of rigid bodies subject to anisotropic motions and
necessary but may be useful if indeed all copies TLS describes this with very little expense in terms
of each molecule are identical. Some double con- of additional parameters. From the optimizational
formations of side chains can still be modelled point of view a single TLS body is as ’expensive’
if clearly visible. Solvent atoms can be modelled for the minimizer as one alanine residue. NCS
automatically but may require visual inspection. restraints must be used in this resolution range if
Medium resolution (between 2.4 and 2.7 Å): available, unless the NCS related copies are clearly
Automated model building with RESOLVE or different. A few well-ordered solvent molecules
ARP/wARP can work for good phase sets and in couldbemodelledandmightevenbebettervisible
many cases RESOLVE can provide more complete than some side chains.
models. However, feeding the RESOLVE model Low resolution (below 3.3 Å): Automated model
as input to ARP/wARP can sometimes produce a building is unlikely to produce anything useful,
model, which is better that a model produced by unless the experimental maps are of outstand-
any of the two programs on their own. Procedures ing quality. Refinement with CNS using a torsion
that incorporate directly anomalous data to itera- angle parameterization would be a first choice.
tive model building and refinement, as described Phase restraints must be exploited. NCS restraints
by, for example, Skubak et al. (2004) can be used must be used if available and, in fact, they will
when available. Refinement can still best be car- offer the only realistic chance for the success-
ried out with REFMAC or PHENIX using isotropic ful refinement at low resolution. Although one
ADPs. BUSTER can offer significant improve- should always try hard to get the best possi-
ments, especially with models that have missing ble model, inability to reduce the free R factor
regions. Phase restraints can be useful in pro- below 35% will not necessarily imply a failure,
viding additional ’observations’ and helping the especially if the model is strongly supported
optimizer find a chemically plausible minimum. by high quality experimental electron density
The use of TLS can still be recommended and NCS and excellent stereochemistry as judged by the
restraints are very likely to be useful. Solvent can Ramachandran plot.
still be modelled automatically but may require
critical inspection. 11.4.2 Model building and refinement starting
Medium-to-low resolution (between 2.7 and 3.3 Å): from an available model
AutomatedmodelbuildingwithRESOLVEislikely
to give some partial models. BUCANNEER — a The discussion here is limited to model building;
novel software from the CCP4 suite — can also advice on refinement can be found in the corre-
be tried. In reality, most of the model will have sponding paragraphs of the preceding section. A
to be built manually using interactive graphics more detailed overview is given in Perrakis et al.
software. Refinement with REFMAC, PHENIX, or (2001) and Carson (2006).
BUSTERusingisotropicADPsshouldperformwell The rule of thumb is relatively simple: the higher
inmostcases. However, refinementinCNS(which the resolution of the X-ray data, the poorer (less sim-
should be also available in PHENIX by the time ilar, less complete) the starting model can be for the
this text reaches the press) using a torsion angle model building to still succeed. With a resolution of
parameterization (Rice and Brunger, 1994) should around 2.0 Åor higher ARP/wARP probably offers
be tried. This way of describing the freedom of the the best chances for success. Initial models having
molecular model requires a much smaller number only 20% sequence identity (Weichenrieder et al.,
of parameters and should therefore improve the 2004) or encompassing only 60% of the scattering
performance of most optimizers. Phase restraints mass (Pichler et al., 2005) have proved sufficient to
are very likely to be useful and should definitely provide an adequate starting point. When the res-
be exploited, regardless of which parameteriza- olution is lower than 2.0 Åor the model is less than
tion has been chosen. The use of TLS can still be about two-thirds complete, automated model build-
helpful. Macromolecules are viewed as collections ing should better be preceded by either statistical