RE: [ccp4bb]: refinement weights

[Kostrewa Dirk <dirk.kostrewa@psi.ch>]

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

In my experience I get very good results in all different refinement
programs when I set the relative weights such that the final rmsd for bond
lengths is about 0.012-0.015 Angstrom and the rmsd for bond angles is around
1.5-2.0 degrees. Regarding the doubling of the wa in CNS: if you look at the
scalenbulk module, which determines that weight, the refined wa is divided
by a factor of 2. This factor was introduced later (as Paul Adams told me),
but I always remove it, because otherwise CNS refines the model with, in my
experience, too tightly restrained geometry.
In REFMAC use the command "WEIG EXPE MATR" followed by a value that has to
be lower for tighter geometry. I usually end up with values around "0.5".

Regards,

Dirk.


*****************************************************
Dirk Kostrewa
Paul Scherrer Institut   e-mail: dirk.kostrewa@psi.ch
Life Sciences             phone: +41-56-310-4722
OSRA/007                    fax: +41-56-310-4556
CH-5232 Villigen PSI        WWW: http://sb.web.psi.ch
Switzerland
*****************************************************


> -----Original Message-----
> From: Bernard Santarsiero [mailto:bds@uic.edu]
> Sent: Mittwoch, 23. Januar 2002 01:40
> To: ccp4bb@dl.ac.uk
> Subject: [ccp4bb]: refinement weights
> 
> 
> ***  For details on how to be removed from this list visit the  ***
> ***          CCP4 home page http://www.ccp4.ac.uk         ***
> 
> While we're on the subject of refinement, with REFMAC, CNS, TNT, and
> SHELX, what do people do to adjust the relative weights.  
> 
> In CNS I typically refine with "wa=-1," which sets up a reasonable
> relative weight between the geometry terms and the xray terms for an
> incomplete model.  Towards the end of refinement, I double or triple
> the value found with "wa=-1." I've found that with "wa=-1" various
> programs like whatcheck and procheck find that the restraints are too
> tight.  When I increase wa, then both the R(cryst) and 
> R(free) decrease a
> bit; at some point R(free) doesn't drop anymore, even if 
> R(cryst) does, so
> that's a reasonable place to stop.  I also look at the rmsd on bond
> lengths (0.01 or lower as a target), and on bond angles (1.5 
> or lower as a
> target).  
> 
> In SHELX, I leave the weight as 0.2, and never mess with it, even with
> small molecule structures.  The sigma's of the intensities seem to be
> correct from various packages (denzo, bioteX, mosflm), since the small
> molecule structures gives GOF's close to 1.0 without the 
> additional fudge
> (weight adjustment) in SHELX.  Since I don't know what the systematic
> errors are in the model or the data, I don't believe that the 
> GOF's should
> be close to 1.0, unless you know you have excellent data and 
> an excellent
> model.  
> 
> I haven't used REFMAC or TNT in awhile, so I'm not sure how 
> the weighting
> schemes are adjusted.
> 
> Bernie Santarsiero
>