Re: [ccp4bb]: Filtering outliers in Patterson maps

[Clemens Vonrhein <vonrhein@globalphasing.com>]

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

Dear Francisco,

I would always run SCALEIT on your data to look at some of it's
output. Even if you have only a single (SAD) dataset: just define
FP/SIGFP and FPH1/SIGFPH1 identical. You don't even need to do the
scaling with that program: you can do analysis only.

The most valuable part I find is the list of reflections that SCALEIT
regards as outliers. It has a little line "===" under the data item
that looks suspicious. Things to look out for:

  - very different FP/FPH1 values, especially at low resolution: maybe
    the beamstop definition during data processing of one of the two
    datasets was not defined properly? Are the resolution limits you
    use really sensible?

  - DPH1 =~ 2*FPH1: If e.g. I(+) was well measured with an intensity
    of 1000 but I(-) was not well measured (beamstop, ice rings,
    cosmic ray ...) with an intensity of 10 (or even negative) - you
    get a huge anomalous difference and a small F. Both are nonsense
    of course.

    This seems to happen quite frequently with SCALEPACK processed
    data (I think, SCALA has E-value based outlier rejection criteria
    which seems to handle things a bit better).

Have a look at the analysis vs. resolution: for isomorphous/dispersive
differences, does the <diso> goes down nicely with resolution? Or is
there a point where it starts going up again (I would start
calculating Pattersons at this resolution limit)?


You could also calculate E values from your differences (FPH-FP or
DANO) and reject very large E values in calculating Pattersons.


In SHARP, we look at the likelihood for each reflection and flag
outliers with very low log-likelihood values. If there are some
obvious outliers at low resolution reconsider your data processing
(beamstop etc.).


There are a lot of places where some statistics are presented
and I would always try to relate these to your own gut about
the quality of your data. If your crystals were bad, data collection
difficult and the crystal died half way through: be surprised if the
programs give you marvelous statistics. On the other hand, if your
crystals are great, data collection went like heaven: the statistics
should be of a similar quality.

Anyway, just my random collection of thoughts ...

Cheers

Clemens


On Fri, Jul 06, 2001 at 09:02:10AM +0100, Francisco J. Enguita wrote:
> ***  For details on how to be removed from this list visit the  ***
> ***          CCP4 home page http://www.ccp4.ac.uk         ***
> 
> Dear Xtallographers :
> 
> Regarding the comments of E. Dodson yesterday about anomalous difference
> Patterson maps, I have a question, perhaps too much basic.
> 
> Is there any general rule to eliminate outliers when we are calculating
> difference (anomalous or isomorphous) Patterson maps for heavy atom
> localization ?. 
> 
> I would appreciate any comments.
> 
> Best regards
> 
> Francisco
> 
> 
> 
> ============================================
> 
> Francisco J. Enguita, Ph.D.
> Protein Crystallography laboratory
> Instituto de Tecnologia Química e Biológica
> Ap. 127
> 2781-901 Oeiras
> PORTUGAL
> Phone : 351-21-4469669
> Fax : 351-21-4411277
> E-mail : fenguita@itqb.unl.pt
> 
> ============================================

-- 

***************************************************************
* Clemens Vonrhein, Ph.D.          vonrhein@GlobalPhasing.com
*
*  Global Phasing Ltd.
*  Sheraton House, Castle Park     Tel: +44-(0)1223-353033
*  Cambridge CB3 0AX, UK           Fax: +44-(0)1223-366889
*--------------------------------------------------------------
* BUSTER Development Group      (http://www.globalphasing.com)
***************************************************************