[phenixbb] Data processing: exhaustive search?

Kay Diederichs kay.diederichs at uni-konstanz.de
Sat Apr 16 13:49:49 PDT 2016


Hi Kaushik,

there is no need for, and no sense in, an exhaustive search of data
processing parameter space, because its dimensionality is way too high
to achieve anything useful. The programs MOSFLM and XDS have undergone
decades of development, and their processing defaults are carefully
chosen. It is rather the user of the programs who is in charge to
correctly interpret the output (which requires reading of their
documentation, and some experience), and to make meaningful adjustments
to (usually very) few parameters.

Speaking for XDS, I recommend to read XDSwiki articles (there are
tutorials and example data sets, and lots of explanations) and a book
chapter that I recently wrote (#131 at
http://cms.uni-konstanz.de/strucbio/diederichs-group/publications ). If
you use the generate_XDS.INP script from XDSwiki, and follow the paper
and the "Optimization" article, you will get good data from XDS. (There
are other scripts, like autoPROC, xia2, xdsme and so on, which usually
also give good results but I don't know them well.) In particular, in
case of XDS there is usually no need to fiddle around with "box size,
spot separation, tolerance and other parameters".

The fact that you seem to try and optimize Rmerge suggests to me that
you are trading precision for accuracy. Too many people still do this,
but it is a legacy of the past. "Trying a combination of better frames
over poorer frames" is most likely _not_ going to improve your merged
intensities - this is another unfortunate and very common
misunderstanding (see references 113 and 130).

I hope this helps you to get better data.

Kay


On 16.04.2016 21:18, phenixbb-request at phenix-online.org wrote:
> Message: 6
> Date: Sun, 17 Apr 2016 00:48:31 +0530
> From: Kaushik Hatti <hskaushik at gmail.com>
> To: PHENIX user mailing list <phenixbb at phenix-online.org>
> Subject: [phenixbb] Data processing: exhaustive search?
> Message-ID:
> 	<CAGMFGb=B9s-j4yXmRx8_M7NPndf86puviiY9QMg=EAqaWr9thw at mail.gmail.com>
> Content-Type: text/plain; charset="utf-8"
> 
> Hello,
> 
> I have a data set diffracted to 2.7A collected at 1.54 wavelength. The data
> is not of great quality with very close spots, ellipsoid spots in certain
> regions of the frames and spot overlaps.  Also certain frames are poorer
> than rest.  The crystal could not be reproduced.
> 
> The best R merge (overall) I have achieved so far is 17% with 85%
> completeness and 3 I/SigI processed in P4 space group. I feel it's possible
> to process it better by identifying right values for box size, spot
> separation, tolerance and other parameters. I am also suspecting a higher
> symmetry space group.  I believe, trying a combination of better frames
> over poorer ones with improve merging statistics.
> 
> I have so far tried processing in iMosflm and XDS. Is there a tool which
> could search exhaustively trying different values for parameters and
> suggest the combination which provide best merging statistics?
> 
> Sorry if this question is not relevant in this group.
> Any pointers/suggestions would be gratefully helpful.
> 
> Thanks in advance,
> Regards,
> Kaushik
> Molecular Biophysics Unit,
> Indian Institute of Science,
> Bangalore, India.

-- 
Kay Diederichs               http://strucbio.biologie.uni-konstanz.de
email: Kay.Diederichs at uni-konstanz.de Tel +49 7531 88 4049 Fax 3183
Fachbereich Biologie, Universität Konstanz, Box 647, D-78457 Konstanz


More information about the phenixbb mailing list