I'd like to add that the value of a molecular replacement solution tends
to be inversely correlated with the effort needed to find the solution.
In other words, the harder you have to work to find the MR solution the
less informative the phase information you tend to get. When you have
very high resolution and/or NCS you may still be able to solve the
structure. However, in cases were the search model is only distantly
related to the protein of interest and Phaser can't find the solution,
the solution may not be worth finding and you're better of focussing on
getting experimental phases.
Bart
Randy J. Read wrote:
On Jan 22 2007, Eaton Lattman wrote:
Will someone knowledgeable tell me what the present state of full 6
dimensional searches in molecular replacement?
Presumably you're referring to systematic 6D searches, not stochastic
ones like in EPMR or QoS. Do you mean "can it be done on current
hardware" or "is it worth doing"? If the former, then it's doable,
though slow. In Phaser, for instance, you can generate a complete list
of rotations (using the fast rotation function with keywords to
prevent clustering and to save all solutions), then feed that big list
of rotations to the fast translation search. In a typical problem that
would probably run on a single processor in significantly less time
than the average PhD, and could be made reasonably quick with a cluster.
If the latter, our feeling is that it isn't worth it. We've tried the
full search option on a couple of monoclinic problems (where it's only
a 5D search), and nothing came up with the full list of orientations
that didn't come up with the first hundred or so orientations.
We conclude that, even in the most recalcitrant cases, the rotation
search gives a better than random indication of whether an orientation
is correct, so it's not necessary to search through all possible
orientations. However, we do feel that it can be worthwhile to try a
reasonably large number of orientations in difficult cases.
Best regards,
Randy Read
P.S. When we generate our list of orientations, we use "Lattman"
angles to get reasonably even sampling of rotations.
