Page 114 - Macromolecular Crystallography

MOLECULAR REPLACEMENT TECHNIQUES 103

also contains the translation function (Storoni et al., 7.7 Non-crystallographic symmetry
2004). Roughly speaking, the rotation score is now protocols and six-dimensional search
based on a weighted origin-removed convolution programs
of experimental and calculated Patterson functions.
If there is NCS in the crystal, all molecules of the
However, the authors stress that their program
asymmetric unit must be searched in turn; every
does not work very well in cases where pseudo-
time a potential solution has been found, it is
translational symmetry is present. This is because
possible to use this information to increase the
the formalism assumes that the translation vector
signal-to-noise ratio of the searches for the other
between the molecules in the unit cell is randomly
molecules. But then, the combinatorics of testing the
distributed so that the relative phases between pairs
50 top solutions of the rotation function and then
of symmetry-related molecules are sampled ran-
the 10 top solutions of each associated translation
domly. This assumption is obviously violated if
function for rigid-body refinement cannot be done
there is pseudo-translational symmetry; however,
‘by hand’ as in the previous case, as soon as there
the authors mention that they have found a way
is more than two molecules in the asymmetric unit.
around this problem and that it will be fixed soon.
In NCS-MR, depending on the number of molecules
Once released, the programers received imme-
present in the asymmetric unit, there are thousands
diate praise from the crystallographic community,
of possibilities to be searched. Also, as one is search-
with messages of thanks posted to the CCP4 Bulletin
ing with only a fraction of the asymmetric unit, the
Board, reporting how several structures that had
signal to be expected is intrinsically lower.
resisted molecular replacement traditional methods
AmoRe (Navaza, 2001), and other programs such
for years were solved in less than a day by PHASER.
as MolRep (Vagin and Teplyakov, 1997, 2000) and, in
So, this is certainly one the packages to try first when
fact, most MR programs handle this quite effectively
dealing with a tough MR problem.
in an expert fashion. MolRep is a very versatile pro-
gram that has many different options implemented
and is part of CCP4 (Collaborative Computational
7.6 Screening many solutions (multi Project, 1994). Obviously, if one wants to try differ-
sunt vocati, pauci vero electi) ent possible models to increase the chance to find the
correct solution, this again makes the search more
The need for automated protocols is apparent from
computerintensiveandthebestwaytodealwiththis
the strategy adopted by AMoRe to circumvent the
is to follow a given, sensible protocol (see below).
problem that the score of the rotation function (RF)
As we mentioned already, the problem of the
is far from being perfect and does not always rank
rotation function is its score, leading to a diffi-
the solutions correctly (Navaza, 2001). Indeed, it is
cult energy landscape to be searched; we can now
often observed that the true solution is not the top
describe another way to tackle this problem. Since
solution, with many false positives. Hence, AMoRe
the Translation Function score is much more sen-
runs a translation function (TF) for each of, typically,
sitive, one might try to run a translation for every
the top 50 or 100 solutions of the rotation function.
possible rotation angle, therefore exploring the 6D
This is actually quite rapid as TF is based on FFT;
space exhaustively. The space to be searched in eule-
then, the first 10 solutions of each of these TF runs is
rian angles depends on the space group of the crystal
in turn refined using a very effective implementation
and can be found in Rao et al. (1980). It turns out that
of rigid-body refinement (Navaza, 2001).
it is doable in most cases within reasonable cpu time
One complication occurs with polar space groups,
with a ‘normal’ workstation.
where all the possibilities must be tried in the TF.
There are at least two implementations of this
If there is an ambiguity in the extinctions, for
protocol that appeared recently:
instance in the Laue group Pmmm, again all pos-
sibilities must be searched. This is usually done ‘by
hand’, going through all different possibilities one • SoMore, which performs a full 6D search with
by one. low-resolution data (usually 8 Ångstrom), followed