I would agree on both points- as to the NCS, there was a very useful study by Kleywegt
of deposited structures that that released NCS at resolutions that didn't justify it-
maybe
G J Kleywegt, "Use of non-crystallographic symmetry in protein structure refinement", Acta 
Crystallographica, D52, 842-857 (1996).
or
G J Kleywegt & T A Jones, "Where freedom is given, liberties are taken", Structure, 3, 
535-540 (1995).

As to outliers, I understand adding (riding) hydrogens can improve the Ramachandran,
but like enforcing dihedrals this somewhat invalidates the use of the Ramachandran,
since it's mainly the clashes between hydrogens that define the ramachandran
permitted area. I would not recommend adding H's at this resolution.

If the outlier is well defined in the density, it is probably a real outlier;
If as you say poorly defined, the data is not strong enough to determine the
rotamer so it drifts out, and the rama plot is correctly indicating this.

Ed

Thomas Womack wrote:
> On 27 Aug 2010, at 10:55, Petr Kolenko wrote:
>
>> Dear crystallographers,
>>
>> I have a structure at 3.3A resolution, 16 identical chains in AU, merohedral twinning present. I started to refine using NCS restraints with chain A as a reference chain. Current Rwork/Rfree is 21/25. There is almost nothing to refine manually in whole structure now. But, refinement without NCS restraints results in Rwork/Rfree of about 17/28. What should I do? Or is it possible to deposit the structure refined using NCS restraints in final refinement?
>
> This seems like a really well-done NCS refinement; using the multi-fold NCS is allowing you to get what is an excellent Rfree and Rwork-Rfree gap for 3.3A data.  Definitely deposit the NCS-restrained version; refining without NCS restraints just increases the number of parameters by a factor sixteen and spends most of those on fitting noise.
>
> 1% Ramachandran outliers at 3.3A also seems entirely reasonable.
>
> Tom
>