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

Re: [ccp4bb]: translational NCS



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

On Fri, 6 Oct 2000, Holly Jing wrote:

> ***  For details on how to be removed from this list visit the  ***
> ***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***
> 
> I have a pseudo translational NCS which relates the 4 mols
> in the asymmetric unit by ~(0.5, 0, 0) and ~(0.25, 0.25, 0). 
> The 4 mols differ slightly on the domain angle as a 
> result of lattice packing. The resolution is 2.4 A, 
> and the space group is C2. MAD maps etc were not that great,
> but I managed to build the model manually, and have refined
> a couple of cycles so far. The 2fo-fc and fo-fc maps have  
> showed many new and nice features, which is encouraging, 
> but R/Rf is about 41/45.  
> 
> Before going further with the traditional procedure, 
> I think I should understand a few things:
> 
> - Instead of selecting R-free set randomly, do I need to
>   select it specially to avoid any "correlation" of 
>   reflections? Some papers suggest to use thin shell
>   selection in the presence of NCS. Is it mainly for 
>   rotational NCS?

Yes it only is relevant for rotational NCS.

> But I have alternative layers of 
>   strong (25%), weak (50%), and very weak (25%) data.   
>   To make it simpler to think, if I had just (0.5, 0, 0)
>   translation, can I consider (2n, k, l) and (2n+1, k, l)
>   as "correlated"?

No these reflections are not correlated at all (at least not more so than
normal neighbouring reflections). Their main difference is that for the even
reflections the diffraction of the molecules related by the (0.5, 0, 0)
translation add up nearly in phase and for the odd ones they nearly cancel
out.

> On the other hand, this translation can
>   also be treated as a pseudo 2-fold rotation if the 
>   crystallographic symmetry is considered, and this pseudo
>   2-fold is parallel with the crystallographic 2-fold.

That is correct, but since the NCS 2-fold is nearly parallel to the
crystallographic 2-fold the "NCS" correlations introduced in reciprocal space
are between reflections related by 2-fold crystallography or nearly so. Since
we only refine against the "unique data", these 2-fold related reflections
will not be in the unique asymmetric unit and will therefore not be in your
working set of refletions.

>   So, is it ok to use thin shell then? Or for every (2n, k, l)
>   reflection with free flag, make the corresponding (2n+1, k, l)
>   also free? What if I have a combination of rotational and
>   translational NCS, or a translation that cannot be 
>   turned into a rotation? 

Selecting your test set is unfortunately not your biggest problem. Just use a
random selection. 

> - What are people's generall experiences in the effects of
>   the test set selection on the outcomes of the refinement? 

NCS correlations lead to a slight model bias in the test set. The extend of
the effect depends on the level of NCS. E.g. I wouldn't worry about 2 or
perhaps even 3-fold NCS, above that it may become significant. The general
feeling is that even with NCS correlations, the R-free will still indicate if
your refinement is heading in the right direction. However, Sigmaa estimation
is also based on your Rfree set. The NCS-induced bias will suggest that the
model is better than it really is leading to improper weighting. So in
theory, yes you may have to worry about high NCS. The problem is that
selecting reflections in thin shells isn't a great solution either. To get a
reasonable number of shells they have to be very thin. As a result a
reflection in a thin shell often does not have its NCS-related reflections in
the same shell, defeating the purpose. Selecting small NCS-related volumes for
Rfree may be better and I have started implementing that in SFTOOLS but never
got to properly testing it, so it is not in the CCP4 version of the program.
In short, I don't think there is an ideal answer. The positive side is, the
higher the NCS, the less the risk of overfitting. (but remember that in your
case of translational NCS this whole issue is not relevant).

> - At 2.4 A, is it now possible to do automatic rebuilding
>   with wARP/REFMAC? How about if many reflections are weak?

My guess is that it isn't going to work unless the warpers have made great
improvements recently.

> - When a large fraction of reflections is weak across all 
>   resolution ranges, would the MAD map quality be largely reduced 
>   because the weak Friedal pairs may not be accurately measured?

Well it is not going to help you. However, the contribution of a reflection to
the map is proportional to the reflection amplitude. So the quality of the map
will be dominated by the 50% of strong reflections.

> Thanks for any suggestions and comments,
> 
> Holly Jing
> 
> P.S. This mighe be the third tough translational NCS case 
>      discussed recently.

There's two more cases here in Edmonton to keep our lives
miserable/interesting.

I am hoping to tweak the crystals into an exact rather than pseudo space 
group by playing with crystallisation conditions.

Bart

===============================================================================

Dept. of Medical Microbiology & Immunology
University of Alberta
1-15 Medical Sciences Building
Edmonton, Alberta, T6G 2H7, Canada
phone:	1-780-492-0042
fax:	1-780-492-7521

===============================================================================