On 15/09/14 18:05, Dale Tronrud wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
>
>
> On 9/15/2014 9:42 AM, Andrew Leslie wrote:
>> Dear Huw,
>>
>> Many thanks for this information. I had seen the notice about
>> Molprobity using CCTBX validation metrics, but I mistakenly thought
>> that Phenix used the CCP4 monomer library, which I think was true
>> once but is clearly no longer the case.
>>
>> Given the very small sigma values associated with the Phenix C4-C5
>> bond length, it could be said to differ by 10.4 standard deviations
>> from the CCP4 monomer library value !
>>
>> Many thanks to all others who have responded too.
>>
>> In response to Dale's comment, the frustrating thing is that this
>> is not interesting at all, it is merely confusing. The resolution
>> of the new structure is only ~3 Å, so there is no way that one can
>> make meaningful statements about distortions from standard
>> geometry. The warnings merely highlight the fact that we do not
>> have adequate library entries.

One wonders what is meant by "we" here? LMB? CCP4 users? Macromolecular 
crystallographers (that also use Phenix)?  Each of these constituencies 
would give a different response...

>>
>     Which is interesting.  ;-)
>
>     To be serious, I do think this is interesting.  Apparently someone
> looking at this molecule has decided that this bond should be 1.49 A
> with a sigma of 0.02 A while someone else looking at the same molecule
> concluded it should be 1.386 A with sigma 0.01 A.

Which is strange.  I'd imagine that the refmac monomer library was 
constructed by libcheck. The input bond order is "double" - but a decent 
dictionary generating program would determine that this is actually 
aromatic - and thus should pick up the distance for an aromatic 
CR56-CR56, i.e. 1.38 (0.02)A. In this case, this has not happened and 
(it seems to me) that the value for a single bond has been used instead, 
i.e. 1.49 (0.02)A.


> The most
> significant source of uncertainty in this target value (whether it is
> a double bond or an aromatic bond) has been ignored.  Each library is
> constructed ignoring the possibility of error in categorizing the type
> of bond.

It seems so.  But is that so bad?  In this case, is there really any 
doubt that this bond order should be "aromatic"?

>
>     I think the lesson is that the sigmas are being seriously under
> estimated when the target values are not derived from real high
> resolution structures.

I would have said the opposite - the Refmac monomer library usually 
overestimates the esds compared to the distributions from small molecule 
crystal structures (I am presuming that the bond orders are unambiguous).

>
>     Isn't it also interesting that the double bond of ccp4 is longer
> than the aromatic bond of phenix?  Not what I would have expected.

Nor I. I suspect a bug in libcheck.

For the record, the results from pyrogen are:

  ANP   C5      C4    aromatic   1.38494    0.0198269 # pyrogen
c.f
  ANP   C5      C4    aromatic   1.388        0.011         # grade

  ANP   C5      C4      N3          126.78    1.33586        # pyrogen
c.f.
  ANP   C5      C4      C3          126.8      0.7                # grade


(I could trim the sfs a bit, I suppose, hmmm...)

Paul.