Page 175 - Macromolecular Crystallography

164 MACROMOLECULAR CRYS TALLOGRAPHY

atoms are initially simply diffracting balls of density overlapping occupancies, etc.), its power for lower
without necessarily having a meaningful chemical resolution should not be underestimated.
identity or connectivity.
One of the most popular refinement programs 11.3.2 Automation in model building
is the state-of-the-art package Refmac (Murshudov
et al., 1997). Refmac uses atomic parameters (xyz, To review all existing model building concepts
B, occ) but also offers optimization of TLS and and programs is beyond the scope of this chapter.
anisotropic displacement parameters. The objective Instead, a small, selected number of key ideas and
function is a maximum likelihood derived residual methods will be mentioned. The current software
that is available for structure factor amplitudes but packages have anyway much in common so pro-
can also include experimental phase information. grams that are not described here should neverthe-
Refmac boasts a sparse-matrix approximation to the less be comprehensible with minimal extrapolation.
normal matrix and also full matrix calculation. The Intuitively, the approaches generally try to mimic
program is extremely fast, very robust, and is cap- what an experienced crystallographer would do.
able of delivering excellent results over a wide range Therefore, many of the methods follow the algo-
of resolutions. rithms originally employed in molecular-graphics
Another highly popular package is CNS (Brunger packages such as O (Jones et al., 1991) or QUANTA
et al., 1998). It offers a wide range of capabilities (Accelrys Inc.) to aid the manual model building
common to other packages and also includes torsion process. All techniques that address automation
angle refinement, which is particularly powerful in macromolecular model building are, to a larger
and useful in the low resolution regime. CNS also or smaller extent, based on the pattern-recognition
uses likelihood targets and the restraints are for- aspect in the interpretation of crystallographic
mulated via energy functions and force fields. CNS electron density map.
employs a conjugate-gradient based optimizer com- Apopularmethodtorepresentanelectron-density
bined with simulated annealing. XPLOR (Brunger, distribution in a way that captures the connectiv-
1993) and its commercial reincarnations are based on ity of the map was skeletonization, proposed back in
similar principles. The PHENIX (Adams et al., 2002) 1974 (Greer, 1974) but put into practice by Jones and
package, which encompasses a dedicated refine- coworkers in the late 1980s (Jones and Thirup, 1986).
ment module, follows some of the CNS ideas but It is a simple and elegant way to reduce an often
uses other, very advanced minimizers. cluttered 3D electron density map to a set of lines
BUSTER/TNT (Bricogne and Irvin, 1996) is that capture the essential chemistry (connectivity)
another likelihood based refinement package that of the map. The method is an iterative procedure
excels especially in cases in which the model is that removes points from a grid representation of the
still severely incomplete (Blanc et al., 2004; Tronrud electron density map as long as this does not break
et al., 1987). It uses atomic parameters but also has their connectivity. A small set of grid points remains
a novel solvent and missing model envelope func- that can be used to produce a skeleton of the original
tion. The optimization method is a preconditioned density. Despite more sophisticated developments,
conjugate gradient as implemented in the TNT pack- skeletonization remains a powerful and commonly
age (Tronrud et al., 1987) that had a faithful audience employed method and is perhaps still the most
in the pre-likelihood era. widely employed technique for computer-assisted
SHELXL (Sheldrick and Schneider, 1997) is often manual macromolecular model building in experi-
viewed as a refinement program for high-resolution mental electron density maps. Related approaches
data only. Although it undoubtedly offers features that make use of electron density extremes include
needed for that resolution regime (optimization of the core-tracing algorithm (Swanson, 1994) and
anisotropic temperature factors, occupancy refine- molecular scene analysis (Fortier et al., 1997).
ment, full matrix least squares to obtain standard The idea of ESSENS (Kleywegt and Jones, 1997) is
deviations from the inverse Hessian matrix, flexi- to recognize secondary structural templates around
ble definitions for NCS, easiness to describe partially each point in the map by an exhaustive search. This