[ccp4bb]: Re: Scaling low-high resolution datasets

[Johan Turkenburg <jpt@ysbl.york.ac.uk>]

***  For details on how to be removed from this list visit the  ***
***          CCP4 home page http://www.ccp4.ac.uk         ***


Further to Victor's point 1.:

Often you need to give an estimate of the initial scale for one sweep 
relative to the other, otherwise scala tends to struggle a little. 
Scalepack is perhaps somewhat more robust in that respect.
Also, be careful about how much of the low res data you include from the 
_high_ res sweep. You may want to exclude the resolution range that 
suffers most from overloads, whilst making sure there is still 
sufficient overlap.

Johan

Victor Lamzin wrote:

> Dear Mark,
> 
> Well, I can try to contribute to the Friday afternoon discussion.
> 
> 1. Poor scaling of low and high resolution passes. What are precisely
> the problems you have met ? If there is no alternative indexing
> between the low and high-res passes, no severe radiation damage
> and there is a sufficient 1/d^2 overlap between the passes, then
> scaling should be straightforward.

> 
> 2. Whether a missing reflection is better than an incorrectly measured
> reflection is a tricky one. None is good, clearly, but a seriously
> incorrectly measured reflection can indeed be harmful. In both cases
> the mean shell intensity is affected and that is used by a refinement 
> program to estimate B factor, sigmaA, solvent continuum model, etc.
> If any of these affect density map, the arp/warp tracing will suffer.
> 
> With best regards,
> Victor
>  
> "Mark J. van Raaij" wrote:
> 
>>Regarding the question of Ed Berry and the reply of Victor Lamzin, what
>>is the best way to avoid the missing overloads problem?
>>We always (well, often...) collect a low resolution dataset before a
>>high resolution dataset, but find they are not easy to scale with the
>>high resolution dataset, even when the high resolution dataset scales
>>well internally and doesn't appear to have suffered significant
>>radiation damage. We use quite standard Scala scaling procedures. Are
>>there scaling procedures which we could/should implement to make things
>>better?
>>In the end, we usually end up using the high resolution dataset on its
>>own and structure solution programs and arp-warp autobuilding appear to
>>work ok, even if 5-10% of the reflections of the lowest resolution bin
>>(typically 20-5 Angstrom) are missing due to overloads and the Wilson
>>plot "breaking down" at low resolution as a consequence.
>>We do however always take care to process carefully using Mosflm to
>>exclude reflections too near or behind the beamstop and do look quite
>>carefully at spot sizes, cell parameters, mosaic spread etc. Here the
>>complaints Mosflm prints at the end of its log and summary files are
>>very useful. The current fashion of interactive data processing may lead
>>to less care being taken.
>>Perhaps a lower average intensity at low resolution in the Wilson plot
>>is less bad if it is the result of missing reflections, rather than
>>wrongly measured reflections? Maximum likelihood should estimate missing
>>intensities more or less correctly. However, if a reflection is meaured
>>wrongly, maximum likelihood will try to fit it in, leading to much
>>greater errors.
>>PS Sorry for this Friday afternoon ramble, but I hope it will lead to
>>some discussion and some learning moments for me...
>>--
>>Mark van Raaij, m.vanraaij@chem.leidenuniv.nl
>>Leiden University, Gorlaeus Laboratories
>>PO Box 9502, 2300 RA Leiden, Netherlands
>>tel. +31 71 527 4414, fax +31 71 527 4357
>>
> 
> 


-- 
PLEASE NOTE CHANGE IN PHONE/FAX NUMBER AND POST CODE

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Dr. Johan P. Turkenburg                     X-ray facilities manager
York Structural Biology Laboratory	
Chemistry Department                        Phone (+) 44 1904 328253
University of York                              Fax   (+) 44 1904 328266
York YO10 5YW   UK                       Email: jpt@ysbl.york.ac.uk
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++