We seem to be back in an old debate - does good geometry ( ie fitting 
predictive models) nrecessarily mean your structure is correct!  If you 
have poor data, and you impose strict geometric checks your structure is 
still unreliable..

As the data gets better you can relax the geometric restraints - you 
will almpost certainly get MORE outliers, not less, but the information 
content of your structure solution is greater.

Years ago I was part of an EU validation group looking at this at we 
wrote a conglomerate paper which examined the validation tools:

(2)   TI: Who checks the checkers? Four validation tools applied to eight
          atomic resolution structures
      AU: Wilson_KS, Butterworth_S, Dauter_Z, Lamzin_VS, Walsh_M, Wodak_S,
          Pontius_J, Richelle_J, Vaguine_A, Sander_C, Hooft_RWW, Vriend_G,
          Thornton_JM, Laskowski_RA, MacArthur_MW, Dodson_EJ, Murshudov_G,
          Oldfield_TJ, Kaptein_R, Rullmann_JAC
      JN: JOURNAL OF MOLECULAR BIOLOGY, 1998, Vol.276, No.2, pp.417-436


The molprobibity statistics are much more complete and accurate than the 
ones we used, but you still have this problem -  wonderful geometric 
stats do not necessarily mean you have a wonderful structure..

Of course really bad geometry - lots of clashes, omega anges of 90 
degrees etc are indicators of problems which need fixing..
  That is no help to you Katja I realise!
   Sorry Eleanor


Nathaniel Echols wrote:
> On Wed, Oct 21, 2009 at 6:20 AM, Katja Schleider <[log in to unmask]>wrote:
>
>   
>> I solved my first crystalstructure and now want to publish it. But how do I
>> know the structure is ready for publication and deposition in the pdb. We
>> can explain our theory with the structure but which factors I have to regard
>> to publish nothing wrong or bad. Can anybody tell how many outliers are
>> allowed as long as they are in a well defined density?
>>     
>
>
> The Molprobity server suggests the following:
>
> < 0.2% Ramachandran outliers
>   
>> 98% Ramachandran favored
>>     
> < 1% Rotamer outliers
> < 1% of residues with bad bonds
> < 0.5% of residues with bad angles
>
> (and the clashscore should be as low as possible.)
>
> -Nat
>
>