[sharp-discuss] ISYM column

Clemens Vonrhein vonrhein at globalphasing.com
Wed Aug 25 12:53:19 CEST 2010


Dear all,

On Wed, Aug 25, 2010 at 09:21:24AM +0100, Phil Evans wrote:
> ctruncate ought to output ISYM and may do in future when someone gets round to it.
> ctruncate gives some more useful statistics than old truncate.
> In due course I shall probably incorporate ctruncate into my successor program to Scala (working title Aimless), and I'll probably output ISYM (but see point 2  below)
> 
> The longer answers are:-
> 
> 1. ISYM only makes a marginal difference anyway in nearly all cases (though it should be used) 

Yes(ish). To give some background of ISYM: if you store your data as
anomalous difference DANO and mean F, you could have such a situation:

  DANO   F
 ----------
   10   75   ==>  F(+) = 80 AND  F(-) = 70
    -   75   ==>  F(+) = 75 AND  F(-) =  -
                  F(+) =  - AND  F(1) = 75

In the second case (where we know that only one of the Bijvoet pairs
was measured, since ANO has no value), ISYM tells the program if the
mean F was actually F(+) or F(-).

Nowadays, a typical dataset will have the anomalous differences nearly
as complete as the amplitudes, so only very few reflections will have
no DANO but an F. However, in very incomplete datasets this will make
a difference - and if you run SHARP with a collection of incomplete
anomalous data (because your crystals keep dying very quick), having
the correct ISYM flag is definitely a good idea.

> 2. there has been a (slow) move in the CCP4 programs to use F+ & F- (or indeed I+ & I-), rather than Fmean & Dano, which is arguably better, since programs can then decide themselves what to do

Hmm, that might only be the second best option: the program can then
only get DANO as a difference of averages instead of the average of
differences - which are two different things. Imagine the following
set of unmerged measurements (working in F instead of instensities and
ignoring sigmas) ... I hope that table comes through correctly:

 Measurements                                                                 <F(+)> - <F(-)> = Diff   <Diff>
 -------------------------------------------------------------------------------------------------------------
 F(+)=1050 F(-)=922 F(+)=1024                                                 1037.0 - 922.0 = 115.0   128.0 
 F(+)=1070 F(-)=930 F(+)=1061                                                 1065.5 - 930.0 = 135.5   140.0 
 F(+)=1086 F(-)=905 F(+)=1087 F(-)=962 F(+)=1094                              1089.0 - 933.5 = 155.5   153.0 
 F(+)=1055 F(-)=912 F(+)=1063 F(-)=969 F(+)=1082                              1066.7 - 940.5 = 126.2   118.5 
 F(+)=1078 F(-)=924 F(+)=1001 F(-)=958 F(+)=1076 F(-)=900                     1051.7 - 927.3 = 124.4   124.3 
 F(+)=1045 F(-)=977 F(+)=1083 F(-)=960 F(+)=1051 F(-)=998                     1059.7 - 978.3 =  81.4    81.3 
 F(+)=1035 F(-)=967 F(+)=1054 F(-)=959 F(+)=1050 F(-)=967 F(+)=1089           1057.0 - 964.3 =  92.7    82.0 
 F(+)=1049 F(-)=955 F(+)=1047 F(-)=973 F(+)=1043 F(-)=974 F(+)=1035           1043.5 - 967.3 =  76.2    79.0 
 F(+)=1024 F(-)=958 F(+)=1089 F(-)=919 F(+)=1065 F(-)=920 F(+)=1033 F(-)=997  1052.8 - 948.5 = 104.3   104.2 
 F(+)=1024 F(-)=968 F(+)=1096 F(-)=949 F(+)=1088 F(-)=901 F(+)=1019 F(-)=963  1056.8 - 945.2 = 111.6   111.5 

If you first average the F(+) and the F(-) you can then only get

  DANO = <F(+)> - <F(-)>

However, if you look at the unmerged measurements you could e.g. get
the difference of each pair (imagine an inverse beam experiment, or
the crystal aligned along a 2-fold to have Friedel pairs on same
image), I.e.

  DANO = < F(+)-F(-) >

As you can see: you get different answers. And with radiation damage
the second might be more accurate.

Anyway, this is not an issue related to truncate/ctruncate usage or
capability (since they already deal with merged data), but rather
something for the merging step (SCALA or AIMLESS etc). At the moment
we only ever calculate DANO as a difference of averages (right?) - in
which case I agree that one could as well use F(+) and F(-)
only. However, calculating DANO as an average of differences might be
better in certain circumstances and it would be nice to keep the
infrastructure (DANO, SIGDANO, F, SIGF, ISYM) in place to then deal
with it when we have added such a feature to our scaling/merging
programs.

Does that make sense?

> 3. Sharp/autoSharp works on files without ISYM as long as you give a blank assignment

Yes - it will pick one randomly then (better: see above).

Cheers

Clemens

-- 

***************************************************************
* Clemens Vonrhein, Ph.D.     vonrhein AT GlobalPhasing DOT com
*
*  Global Phasing Ltd.
*  Sheraton House, Castle Park 
*  Cambridge CB3 0AX, UK
*--------------------------------------------------------------
* BUSTER Development Group      (http://www.globalphasing.com)
***************************************************************


More information about the sharp-discuss mailing list