[ccp4bb]: summary: what's the best way to combine the phases from varioussources?

[Qian Xu <qian@med-biophg.bu.edu>]

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

Dear all,

Many thanks to all the people who responded to my question "what's the
best way to combine the phases from various sources". Here is a summary of
their suggestions, 

Original question: 

> I have many datasets for a protein from various sources, including
> MAD, SIRAS and MIR from different derivatives. Some of them are
> not isomorphous. I am just wondering whether there is any way by which I
> could refine and phase all of these derivatives in one single run of
> MLPHARE? (One problem is that I can't define different "natives" for
> different datasets, which I believe is necessary). If I can't do that,
> what's the best way to combine all of those phases from various sources?
> I know sigmaa can combine two sets of MIR phases. Is there any other
> program which can do this? and anything I ought to know for optimizing
> phase combination?   

suggestions:
------------
>from rams@poori.biochem.uiowa.edu

The simple easy method that works is just to write out the H-L
coefficients for each individual refinement.  Then simply add them up
using SFTOOLS. The HL coefficients (phases) are very robust and
non-isomorphism bothers them very little.

>from Hay Dvir<hay.dvir@weizmann.ac.il>

You can try to run SOLVE, using the combine script !

>from Ronaldo Alves Pinto Nagem <nagem@lnls.br>

I don't know if it will be easy for you to install SHARP in your computer
but I guess this is one of the best programs available for phasing. You
will just need heavy atom coordinates from each dataset and the datasets
themselves. It will take a while to combine phases from all datasets but I
am almost sure they (phases) will be reliable.

>from Frank Vondelft <Frank.Vondelft@syrrx.com>

SHARP is what you want - it does exactly all that.

>from Bart Hazes <bhazes@ualberta.ca>

If you believe you need different native datasets to combine with your
various derivatives, then you can't come up with one set of phases. Try to
work with different subsets that each are sufficiently isomorphous. If one
subset gives phases of adequate quality your problem is solved. If not,
you could consider multiple crystal averaging across the maps derived from
the various data subsets.

>from Anastassis Perrakis <perrakis@nki.nl>

use dm_multi after sperately phasing your 'natives'.
Since most are more or less isomorphous as I udenstand start
from a unity matrix and let it refine.

>from Eleanor J. Dodson <ccp4@ysbl.york.ac.uk>

 In the end you have to choose a master data set which you want to
phase, and phase that.

 So the way I would proceed.
 
First make sure all our sites are as close to the origin as possible -
that minimises the effect of cell differences.
Then use native1 with derivatives 1H1, 1H2, etc, to refine the 1H1, 1H2,
..... sites and get ISOE1 ANOE1.

Use native2 with derivatives 2H1, 2H2, etc, to refine the 2H1, 2H2, ...
sites.

 You want to use these sites, but get better estimates of ISOE2. ( ANOE2
wont change..

 So I would do one or two cycles of refinement of eac of 2H1, 2H2, etc
against native1, just to get the ISOEs and maybe let shift the
coordinates a bit but not the occupancies.

 Then do a final phasing run with all the derivatives v native1. 
In one case where we had awful non-isomorphism we could only get useful
information to quite low resln for the second set.

An alternative is to just add the HLA1 HLB!.. to HLA2 HLB2.. in Sigmaa
but that takes no account of weighting the non-isomorphism

 I would be interested to hear how it worked. 

 Another way would be to use the two sets for multi-crystal averaging.
See Kevin Cowtans web page for a lecture where he has some discussion of
this.
www.ysbl.york.ac.uk/~cowtan 
------------------------------------------

Thanks again.

regards,
Qian Xu
Dept. of Physiology and Biophysics
Boston Univ. School of Medicine
Tel: (617)638-4083
Fax: (617)638-4041
E-mail: qianxu@bu.edu