[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ccp4bb]: summary: high resolution refinement

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

this is the summary of the responses I recieved to my question about
high resolution refinement. Thanks a lot for all who contributed!!

It seems that different people would prefer different strategies in this
particular case.

Here is once again the question followed by the responses:
I would like to pose one question concering refinement of
high-resolution structures. This topic has been discussed to some extend
at the beginning of this year, but I would like to go for sure and have
an additional question:

The situation:
I have refined a quite large structure (6 monomers of 50kDa each in the
asu) at 1.7AA to quite reasonable R-values (Rwork: 16%, Rfree: 19%). Now
I have measured a 1.295A data set of the same crystal. The unit cell
dimensions differ less than 1%. Because of the huge amount of data
(close to 700.000 independent hkls) refmac5 and ARP/wARP with refmac4
appear to be THE refinement-programs to use (also in terms of tolerable
time for refinement).

The questions:
Is it the reasonable way to start the refinement of the 1.295AA
structure right from the beginning with all the data I have and let
maximum likelihood find its way. Or is it better to use the "old
fashioned" way to extend the high resolution limit in small steps and
search water in each step?

Is the following refinement strategy reasonable (each step until Rfree
- refinement at 1.295AA without water starting from 1.7AA structure with
- adding water with ARP at 1.295AA
- refinement with anisotropic B-factors
- introduction of multiple conformations
- refinement of occupancies
- refinement with hydrogen contributions

Taking in account resolution and the amount of data: Where should the
R-values converge after proper refinement?

Thanks for any comments!                           
The answers:

** Eleanor Dodson **
Various comments ..
 I would always start with a bit of rigid body to accommodate the small
changes in cell dimensions..
 Cant see any reason to throw away hard earned water positions - the
wrong ones should disappear soon enough .
 Biggest challenge and most time is needed to find alternate
conformations, and ARPing can make this harder ...
 My order is different..
 Automatic refinement of complete model against all data( prob. with
hydrogens - cant see why not use something which is chemically sensible)
- rigid body first then isotropic
 Sort refined model on B values, and look at highest 10-15% to see what
has happened..
 You usually pin-point some residues which are in loops - maybe all
copies have the same conformation, and can be rebuilt into an averaged
map; others which are obv. in multiple conformations - usually with H2Os
sitting where the 2nd conf should be..
 Once you have made the obvious corrections then maybe start adding more
waters, anisotropic parameters, etc.
 Protocols can differ to accommodate NCS

** David Borhani **
1. The structure is very likely substantially correct as is. Do some
rigid body
refinement to position it correctly in the new cell, then refine to
at the old resol of 1.7 Ang with the new data. Include in this process
overall B-factor refinement, as the B's have likely gotten much smaller.
maps for alt. confs, clear new waters, etc., and rebuild as needed.  I
would use
the 1.7 Ang water set as a starting point.
2. Push resol. to 1.3 in 2 jumps...you can try aniso B's at ~1.5 Ang,
definitely at 1.3. Also add in the hydrogens. Your R's should drop
at this point (2-5%).
3. I would be interested if others have very different suggestions.

** Victor Lamzin **
Dear Jan,
I suggest to use all data straight away. Since your cell parameters
slightly differ and the position of the molecule may be affected,
the quickest would be either:
Run molecular replacement
Run PDBSET/COORCONV to convert crds to frcs and then back to orths with
the new cell parameters.
In our experience it is quicker to start with addition of solvent sites
at 1.3 A (refinement of both bulk solvent and overall anisotropy within
REFMAC), followed by introduction of riding hydrogens, then anisotropic
refinement and finishing up with multiple conformations and refinement
or estimation of their occupancies.
Very much depends on the Wilson plot B factor and overall anisotropy.
A typical value for R factor would be 11-13 %.
Regards, Victor

** Garib Murshudov **
If you are using refmac5 you don't need to run COORDCONV. refmac5 itself
convert cell dimensions of coordinate file according to mtz.
My experince with higher resolution data refinement agrees with what
Victor is
saying. Use all data straight away. Especially in the cases where you
have refined
data to a little bit lower resolution there should not be a problem.

** Mark van Raaij **
A few small comments:
I would put in multiple confs before individual anisotropic Bs, as the
latter can sometimes mask multiple conformations. Otherwise the
protocol looks fine to me. I don't think there is any point in doing
of higher and higher resolution. the maximum likelyhood combined with
checking on the graphics should sort things out fine.
The target R and Rfree are difficult to estimate it depends a lot also
the quality of the data, it will be important to check the refmac output
carefully for geometry statistics and to do validation for example by
Whatcheck. Then you can vary the geometrical restraints so as to
Whatcheck complaints as well as Rfree...                      


Jan Abendroth
Institut fuer Biochemie
Universitaet Koeln
Zuelpicher Strasse 47
D-50674 Koeln

Tel: +49-(0)221 470 6455 (lab)
                    6445 (office)
Fax: +49-(0)221 470 5092