Page 177 - Macromolecular Crystallography

166 MACROMOLECULAR CRYS TALLOGRAPHY

decade: SOLVE (Terwilliger and Berendzen, 1999) decisions to be made. These decisions involve choice
for experimental phasing of the diffraction data and of the crystallographic program most suitable for
ARP/wARP (Perrakis et al., 1999) for model build- each task, together with optimal input parameters
ing and refinement. Over the last few years, a drive for each of these programs. The important param-
from Structural Genomics initiatives has boosted a eters include the space group of the crystal, the
variety of outstanding activities. Most notable is the number of molecules in the asymmetric unit, the
PHENIX project (Adams et al., 2002) that has recently type of the heavy atom derivative, the extent of
been released. Other automation attempts include a derivatization, the diffraction limit of both the native
series of the so-called software pipelines, for exam- and the derivatized crystal and the quality of the col-
ple autoSHARP (Bricogne et al., 2002; de La Fortelle lected diffraction data. If the data collected in the
and Bricogne, 1997), BnP (Weeks et al., 2005), Elves first experiment during, for example, MAD mea-
(Holton and Alber, 2004), ACrS (Brunzelle et al., surements can be successfully interpreted, further
2003), CRANK (Ness et al., 2004), andAutoRickshaw data collection can be halted (Dauter, 2002). The
(Panjikar et al., 2005). choice of model building step within AutoRickshaw
Commonly known in the Structural Genomics depends on the maximum resolution of the X-ray
jargon as software pipelines, these are systems that data. If it is lower than 2.6 Å, ARP/wARP Version
combine a number of macromolecular crystallo- 6.1 or ESSENS are invoked to identify helices in the
graphic computer programs with several decision- electron density map. For higher resolution ARP/
making steps linking the computational modules. wARP is used for tracing polypeptide chains.
Pipelines can be a simple chain-type sequence of An important aspect in all pipelines is the evolu-
steps or can involve internal loops at several lev- tion of the decision-making. As more data become
els of complexity, exemplifying the inadequacy of available, the structure determination paths can
the term ‘pipeline’ to describe them. In the latter be scrutinized thoroughly in order to increase the
case there could be many different paths through efficiency of the overall workflow.
which the structure determination evolves and these
paths can either be predefined or modified on the fly, 11.4 Model building and refinement
making that software rather complicated decisions cookbook
systems with significant amounts of pre-existing sci-
entific knowledge and new ideas incorporated in Coming up with robust recipes for model building
them. Some of these automated decisions systems and refinement is a challenging task. Automated
rely mainly on one software package while others pipelines attempt to capture as many successful
are more comprehensive. approaches as possible but still fail in some cases.
Below we briefly describe the crystallographic If automation fails, the user has to understand the
software pipelines using AutoRickshaw as an exam- fundamental logic behind the software and grasp,
ple, with its flexibility and the ability to decide on at least qualitatively, the underlying mathematics,
the path to be taken dependent on the outcome of and successfully identify potential problems before
a previous step. On one hand, AutoRickshaw has making informed decisions. Seeking the advice of
features and general steps, which are also shared by an experienced expert crystallographer can rarely
many other pipelines. On the other hand, AutoRick- be avoided. In this section we give some general
shaw is perhaps the first software pipeline which guidelines for situations that an average crystallo-
aims not at the delivery of a fully built, refined, and grapher can encounter. It must be emphasized that
validated model but rather at fast evaluation of the each ’case study’ should be treated critically, espe-
quality of the X-ray data in terms of interpretability cially the resolution margins that characterize each
of the obtained electron density map. case and the reader may be advised to keep in mind
AutoRickshaw considers crystal structure deter- Fig. 11.3 and adjust the conclusions accordingly. It
mination as a multistep process in which each step in should be noted, that in quoting the resolution lim-
structure solution, from substructure determination its we will refer to the highest resolution of at least
to model building and validation, requires certain one dataset, which typically (but not necessarily)