[phenixbb] R free assignment in thin shells

[Pavel Afonine]

Hi Mohamed and Tim,

I'm yet to see if using "thin-shells" vs random make a notable 
difference, however, for the record, I think it is worthwhile to point 
out this paper where authors explain and demonstrate why using 
"thin-shells" is important in case of high-order NCS:

Acta Cryst. (2006). D62, 227-238
Bias in cross-validated free R factors: mitigation of the effects of 
non-crystallographic symmetry
F. Fabiola, A. Korostelev and M. S. Chapman

On ML and Rfree comment:

- ML is superior than LS if your model (atomic + nonatomic) misses 
scattering, that is incomplete, because ML takes missing scattering into 
account statistically;

- free set must be available for ML target calculation, so if you use ML 
then you get Rfree as a bonus.

Pavel



On 6/4/15 12:21 AM, Tim Gruene wrote:
> Dear mohamed noor,
>
> thin shells are commonly picked in cases of twinned data. As Andrew
> points out in the email you link, the NCS makes the refinement behave
> well, so also you will probably get away with 'normal' randomly selected
> reflections.
>
> At high resolution the data to parameter refinement also increases and
> the Rfree looses its meaning. That's one of the reasons why people don't
> use Rfree for small molecule structure determination and why ML
> refinement show not much improvement compared to least squares
> minimisation (it's different for charge density studies, when the data
> to parameter goes down again).
>
> Regards,
> Tim
>
> On 06/04/2015 01:18 AM, mohamed noor wrote:
>> Dear all
>>
>> I found a not-so-old thread about this topic:
>> http://phenix-online.org/pipermail/phenixbb/2012-January/018259.html. Maybe
>> there has been significant changes in phenix.refine since then.
>>
>> However, I couldn't find a conclusion. In my case, my data extends to 1.4 A
>> with the R/R free being around 23/26 %. There is also pseudo symmetry, so
>> there are 16 NCS copies in a lower symmetry group than what Xtriage
>> suggested. The map looks fine with density for some H atoms even being
>> visible.
>>
>> On what basis should I decide whether to use (or not) thin shell
>> assignment? And what is the cut-off for number of NCS copies?
>>
>> Thanks.
>>
>>
>>
>> _______________________________________________
>> phenixbb mailing list
>> phenixbb at phenix-online.org
>> http://phenix-online.org/mailman/listinfo/phenixbb
>> Unsubscribe: phenixbb-leave at phenix-online.org
>>
>
>
> _______________________________________________
> phenixbb mailing list
> phenixbb at phenix-online.org
> http://phenix-online.org/mailman/listinfo/phenixbb
> Unsubscribe: phenixbb-leave at phenix-online.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20150604/a63cd783/attachment.htm>