[ccp4bb]: Re: Data Reduction for mosaic crystals...

[Edward Berry <eaberry@lbl.gov>]

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

I'm also interested to hear what others will say, so I post this for others 
to shoot down or not as they see fit-

I can see two reasons for fine slicing, one that says fine slicing helps 
with very low mosaicity crystals, and the other that says fine slicing is
needed with high mosaicity.

The first has to do with signal to noise: If a reflection is only in the 
diffracting condition for 0.2 degrees and you collect 2 degree oscillations, 
then 90% of the oscillation is only adding noise. Better to collect the 
spot in one or two fine slices, then the integrating program can ignore 
the rest of the 2 degrees (as far as that spot is concerned).

On the other hand if the mosaicity is greater than 1 degree, then the
reflection will be diffracting during most of the time, so even if you 
fine slice the integrating program will have to add together all those 
slices and their associated noise. Furthermore if there is a baseline 
"per-image" noise such as dark current error, it will get added in 
ten times instead of once.

So that argument says you only fine-slice with low mosaicity.
On the other hand if you have an overlap problem resulting from lunes
of the diffraction pattern overlapping (as opposed to spots within a 
lune overlapping due to large cell parameters), then it would be 
reasonable(?) to guess that mosaicity and oscillation angle are 
additive in contributing to the width of the lunes. Say with your 
crystal, at your resolution, overlap becomes a serious problem when 
the sum of the mosaicity and oscillation angle is greater than 
1.2 degrees. If the mosaicity is 0.2, then you can collect 1.0 degree 
oscillations. If the mosaicity is 0.7, you should collect 0.5 degree 
oscillations. You will have no fulls, but you will collect all the
partials and they can be added together. A logical extension (although 
I suspect not a practical one) is that if you have 1.1 degree
mosaicity you should collect 0.1 degree oscillations.

In my experience in dealing with mosaic crystals with large cell parameters 
it is possible, as a last resort, to cheat a little on mosaicity. First 
integrate with a reasonable mosaicity, and postrefine to see if the value
was in fact reasonable. Completeness drops off badly with resolution due to 
overlap. 
   Then run denzo again using 2/3 or even 1/2 the true mosaicity,
and scalepack WITHOUT FITTING MOSAICITY (otherwise scalepack will 
realize he needs a lot of partials that weren't measured). Completeness 
improves greatly, and R-merge (R-sym?) goes from around 6% to around 12%,
partly because we're including more weak high-resolution data but more 
seriously because we're chopping the tails and ignoring the overlapping 
tails of neighbors. A 6% increase in R-merge on I corresponds maybe 
to a 3% increase in R-free on F, and if R-free is on the order of 25-30%
this may not seem so bad. On the other hand the error is systematic, 
won't be reflected in the sigma values you are asking the refinement 
program to trust, so this option is sure to make the purists cringe. 
Still it's something to do while waiting for the perfect post-crystal-growth
treatment to reduce your mosaicity to a fraction of a degree.

Ed

Anastassis Perrakis wrote:
> 
> ***  For details on how to be removed from this list visit the  ***
> ***          CCP4 home page http://www.ccp4.ac.uk         ***
> 
> > Here are all the responses to my questions! Thank you very much.
> > Raji
> >
> > Questions:
> > I am working on a dataset where the crystal has a very high mosaicity - my
> > estimate is ~2.0 or more. I have had to take 0.1deg frames during data
> 
> Since I am a bit dissapointed that I failed to provoke a good discussion
> with my last answer on that one, but I still find it interesting,
> I will rephrase my last mail as direct questions and wait for some lively
> discussion:
> 
> If you DO have crystals with 2.0deg mosaicity:
> 1. Am I alone thinking Raji did *not 'have had'* to take 0.1deg frames ?
>     Besides the good ideas offered for really processing the data,
>     does fine phi slicing improve anything for *very mosaic* crystals ?
> 2. Do people get good maps out of such datasets ?
> 3. Have people solved non MR structures out of >1.5 def mosaicity data ?
> 4. Whats at the end the minimal meangfull phi slice to get as fucntion of
> mosaicity ?
> 5. Any tricks peopel want to share with us for reducing mosaicity (annealing?) ?
> 
>     Tassos
> 
> P.S. Sorry for continuing this non-ccp4-directly related discussion but I really
> think we can
> learn from some comments of people that had similar experiences with very high
> mosaicity - its a problem everybody has seen at some time or the other !
> 
> --
> NKI, Department of Molecular Carcinogenesis
> Plesmanlaan 121 1066 CX Amsterdam NL
> Phone +31-20-512-1951 Fax +31-20-512-1954
> SMS: +31-6-28597791 WWW: http://den.nki.nl