Dear Ian,
On Fri, Sep 12, 2008 at 05:53:35PM +0100, Ian Tickle wrote:
>
> I would have thought that it would always be a good idea to refine only
> the occupancies in the first few cycles and only refine co-ords & B
> factors once the occupancies have settled down to sensible values.
Correct, this is exactly what SHARP/autoSHARP does by default:
refining occupancies and scale factors (including non-isomorphism
parameters) first, then adding coordinates and finally B-factors.
> But in that case wouldn't the Fcalc's be linearly dependent on the
> occupancies so the occupancy-only refinement should be reasonably
> well-behaved, even from a distant starting point? The initial
> guesstimates of the occupancies could be significantly off even if
> the data is correctly scaled so the refinement still needs to be
> reasonably robust.
It all depends how far away 'distant' really is I guess. Imagine a
case where you only have poor 4A data, with a possibility of 6-24
mol/asu (ok, at that resolution a low number is more likely), you
found two heavy atom positions of a soak (so no idea about real
occupancy anyway), have no idea about B-factor (100? 200?) for these
heavy atoms and you want to phase using SIRAS with a fairly
non-isomorphous native (2A cell shift) to go with the derivative (that
is only 85% complete). And for good measure you had severe radiation
damage and highly anisotropic diffraction of a split crystal.
Unfortunately this doesn't sound too unlikely to me I must say
... seen a few of those cases that shout "Grow better crystals!". But
the poor scientist is desperate to finish the thesis/project on time
because the PI/boss/patent lawer is pressuring and the next
position/job/Nobel prize is already sorted ... and the competition is
starting writing up ... etc.
In such a case you have to try to eliminate as much uncertainty as
possible. And having the data on a roughly absolute scale (still a
factor of 10 or more off is fine) might help - even if it only gives
you the positive feeling that at least that part of a difficult
project was dealt with correctly ;-)
> But I agree with you, there can't be any harm in applying the Wilson
> scale factor, so you may as well always do it, at least it ensures that
> the range of values for the F's is roughly right (and you're less likely
> to get questions like "why are my occupancies more than 1 ?"!).
Another argument is that if people write ASCII files containing
intensities/Fs the values will fit into the particular format (one of
the advantages of a binary format like MTZ: you're unlikely to loose
accuracy by hitting some arbitrary printf/write statement).
Cheers
Clemens
--
***************************************************************
* Clemens Vonrhein, Ph.D. vonrhein AT GlobalPhasing DOT com
*
* Global Phasing Ltd.
* Sheraton House, Castle Park
* Cambridge CB3 0AX, UK
*--------------------------------------------------------------
* BUSTER Development Group (http://www.globalphasing.com)
***************************************************************
|