[buster-discuss] Clarification: Not French Wilson

[Dale Tronrud]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


   Much as I'm reluctant to comment on a private letter forwarded to a
public bulletin board I think I will because Peter seems to have made
a misstatement.

   The "Plain" French & Wilson method describes a Bayesian procedure
for converting I's to F's (not scaling at all I'm afraid) where the
prior contains knowledge that negative F's are impossible and positive
F's should obey Wilson statistics.  This means that "massage
intensities" is one step on the road to French & Wilson, not a step
beyond.  Since Wilson statistics does not provide a very informative
prior I don't think there is much difference between the two - but
since Phenix provides both the test is pretty simple.

   When the simple-minded version of Wilson statistics implemented in
Truncate (in CCP4) is wrong (e.g. ansiotropic fall-off,
pseudo-centering) it would be best to leave it out and use the
"massage intensities" method alone.

Dale Tronrud

On 11/25/2014 5:51 AM, esserlo at helix.nih.gov wrote:
> Hi,
> 
> in my previous e-mail (I don't know what behooved me to do so) but
> I mixed up two things that should not be confused as Peter pointed
> out (see below).
> 
> My question simplifies to this:
> 
> Based on the observation that Buster refine expects amplitudes and
> not intensities (I tried but it complained), I have the option to
> covert I to F in the traditional way (here I am using
> phenix.reflection_file_converter using defaults) or I can use the
> option "massage-intensities" which is a process of altering mostly
> weak reflections incl. those with negative intensities using the
> prior "knowledge" that they all should be positive. The first time
> I learned about the formula written down by Peter (|Fnew|) is when
> I read it in the book "Data Analysis A Bayesian Tutorial by D.S. 
> Sivia".
> 
> Well, the questions is this: Which option should I use ?
> 
> It may matter that I am working with a low resolution data set
> which likely has quite a number of weak reflections. If I had a
> very strong data set, it is likely that massage-intensities makes
> no difference whatsoever.
> 
> Thanks,
> 
> Lothar
> 
> 
> Hi Lothar,
> 
> Although I do not know the preferences of BUSTER, I thought it
> would be good to point out that phenix provides two options.
> 
> -1- plain French & Wilson scaling -2- 'massage intensities'
> 
> The first one is an amplitude estimate on the basis of Wilson (a
> different Wilson in this case) statistics. This is equivalent to
> the CCP4 code AFAIK.
> 
> The second option is doing something that is obtained when you
> assume a flat, improper positive prior on the intensities,
> resulting in
> 
> |Fnew| = sqrt((Io+sqrt(Io**2 +2sigma**2))/2.0).
> 
> I derived this myself a while ago and subsequently found that this
> was (of course) done by (I think) David, but cannot find the
> reference at this point in time. The beauty of the method 2 is that
> it is not sensitive to anisotropic scattering and/or pseudo
> centering, which can be useful in some cases.
> 
> I do not know what buster likes, but it is good to know that there
> are several options.
> 
> Given that this is a phenix specific email on a Buster mailing
> list, I have send this to you and Clemens Vonrhein only to avoid
> ruffling feathers.
> 
> Cheers Peter Zwart
> 
> 
> 
> On 24 November 2014 at 13:11, <esserlo at helix.nih.gov> wrote:
> 
>> Hi,
>> 
>> when I convert my data file from phenix.refine which uses
>> intensities to autobuster which requires amplitudes, I have the
>> option to "massage-intensities" which is French and Wilson
>> scaling. My question is whether French and Wilson scaling should
>> be applied or not. Or should I always test both ways and pick the
>> one that gives lower R-free values? This seems like a bit of a
>> trivial question but I could imagine that the buster developers
>> have strong opinions about how data should or should not be 
>> treated.
>> 
>> Thanks,
>> 
>> Lothar
>> 
>> _______________________________________________ buster-discuss
>> mailing list buster-discuss at globalphasing.com 
>> https://www.globalphasing.com/mailman/listinfo/buster-discuss
>> 
> 
> 
> 
> -- 
> ----------------------------------------------------------------- 
> P.H. Zwart Staff Scientist Berkeley Center for Structural Biology,
> Science lead Lawrence Berkeley National Laboratories 1 Cyclotron
> Road, Berkeley, CA-94703, USA Cell: 510 289 9246 SASTBX:
> http://sastbx.als.lbl.gov BCSB:      http://bcsb.als.lbl.gov 
> PHENIX:   http://www.phenix-online.org 
> -----------------------------------------------------------------
> 
> 
> 
> _______________________________________________ buster-discuss
> mailing list buster-discuss at globalphasing.com 
> https://www.globalphasing.com/mailman/listinfo/buster-discuss
> 
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (MingW32)

iEYEARECAAYFAlR0v2QACgkQU5C0gGfAG13R5gCeNoho5M//N9rgHQAXmmXpoyAN
TBUAn0kxXSrjQR1cGt+XIxlf31cLHlq4
=YIgG
-----END PGP SIGNATURE-----