Hi Xiao,

That all looks normal.  If you have processed the data in too low a symmetry (as you deliberately have) then there will be many "pseudo-merohedral" twin possibilities.  The key thing here is the lack of translational pseudo-symmetry and the near-perfect match of the Wilson moments to a perfect twin.  I'd say there is a high probability that this is a nearly-perfectly-twinned crystal.

All the best,
Tom T

On Dec 10, 2014, at 3:45 PM, Xiao Lei wrote:

Hi Tom,

I paste the patterson analysis from the xtriage log file here. This shows no significant pseudotranslation is detected.However in the summary of the log file it gives me some PM (pseudo-merohedral) possiblities.


Patterson analyses
------------------

 Largest Patterson peak with length larger than 15 Angstrom 

 Frac. coord.        :    0.269   -0.000   -0.001
 Distance to origin  :   18.194
 Height (origin=100) :    2.560
 p_value(height)     :    1.000e+00


   The reported p_value has the following meaning:
     The probability that a peak of the specified height
     or larger is found in a Patterson function of a
     macro molecule that does not have any translational
     pseudo symmetry is equal to  1.000e+00.
     p_values smaller than 0.05 might indicate
     weak translational pseudo symmetry, or the self vector of
     a large anomalous scatterer such as Hg, whereas values
     smaller than 1e-3 are a very strong indication for
     the presence of translational pseudo symmetry.


Patterson analyses
  - Largest peak height   : 2.560
   (corresponding p value : 1.00000)


The largest off-origin peak in the Patterson function is 2.56% of the 
height of the origin peak. No significant pseudotranslation is detected.


-------------------------------------------------------------------------------
Twinning and intensity statistics summary (acentric data):

Statistics independent of twin laws
  <I^2>/<I>^2 : 1.554  (untwinned: 2.0, perfect twin: 1.5)
  <F>^2/<F^2> : 0.878  (untwinned: 0.785, perfect twin: 0.885)
  <|E^2-1|>   : 0.557  (untwinned: 0.736, perfect twin: 0.541)
  <|L|>, <L^2>: 0.375, 0.199
  Multivariate Z score L-test: 11.683

 The multivariate Z score is a quality measure of the given
 spread in intensities. Good to reasonable data are expected
 to have a Z score lower than 3.5.
 Large values can indicate twinning, but small values do not
 necessarily exclude it.



Statistics depending on twin laws
------------------------------------------------------------------
| Operator  | type | R obs. | Britton alpha | H alpha | ML alpha |
------------------------------------------------------------------
| h,-l,k+l  |  PM  | 0.359  | 0.127         | 0.118   | 0.022    |
| h,k+l,-k  |  PM  | 0.359  | 0.127         | 0.118   | 0.022    |
| h,-k-l,k  |  PM  | 0.074  | 0.428         | 0.429   | 0.435    |
| h,l,-k-l  |  PM  | 0.074  | 0.428         | 0.429   | 0.435    |
| h,-k,-l   |  PM  | 0.370  | 0.117         | 0.118   | 0.022    |
| -h,k,-k-l |  PM  | 0.361  | 0.124         | 0.118   | 0.022    |
| -h,-k-l,l |  PM  | 0.362  | 0.127         | 0.118   | 0.022    |
| -h,-l,-k  |  PM  | 0.058  | 0.441         | 0.448   | 0.441    |
| -h,l,k    |  PM  | 0.362  | 0.123         | 0.117   | 0.033    |
| -h,-k,k+l |  PM  | 0.075  | 0.427         | 0.428   | 0.452    |
| -h,k+l,-l |  PM  | 0.077  | 0.425         | 0.427   | 0.418    |
------------------------------------------------------------------

Xiao

On Wed, Dec 10, 2014 at 2:17 PM, Terwilliger, Thomas Charles <[email protected]> wrote:
Hi Xiao,
Yes, that is pretty convincing.  One more thing to check:  In your xtriage output what does it say about Patterson peaks? (Is there any translational pseudo-symmetry?)  If not...you're probably done, yes it is nearly perfectly twinned.
All the best,
Tom T

On Dec 10, 2014, at 3:13 PM, Xiao Lei wrote:

Hi Tom,

Thank you very much for your advice, I just tried Xtriage analysis of the data scaled as P1, the Wilson moment is 1.554, this result shows that the dataset is nearly perfect twinned.

I paste the part of the log file of xtriage below.


Analyses of the absences table indicates a number of likely space group
candidates, which are listed below. For each space group, the number of
absent violations are listed under the '+++' column. The number of present
violations (weak reflections) are listed under '---'. The last column is a
likelihood based score for the particular space group.  Note that
enantiomorphic spacegroups will have equal scores. Also, if absences were
removed while processing the data, they will be regarded as missing
information, rather then as enforcing that absence in the space group choices.


-----------------------------------------------------------------------------------
| space group | n absent | <Z>_absent | <Z/sigZ>_absent | +++ | --- | score       |
-----------------------------------------------------------------------------------
| P 1         | 0        |     0.00   |     0.00        |  0  |  0  |  0.000e+00  |
-----------------------------------------------------------------------------------




Wilson ratio and moments 

Acentric reflections 
   <I^2>/<I>^2    :1.554   (untwinned: 2.000; perfect twin 1.500)
   <F>^2/<F^2>    :0.878   (untwinned: 0.785; perfect twin 0.885)
   <|E^2 - 1|>    :0.557   (untwinned: 0.736; perfect twin 0.541)

Xiao

On Wed, Dec 10, 2014 at 1:24 PM, Terwilliger, Thomas Charles <[email protected]> wrote:
Hi Xiao,

Here are some things to check before concluding whether the data are twinned:

1. What are the Wilson moments (2 for untwinned 1.5 for perfect twin)?

2. Is it possible you have the wrong space group?   If the data are overmerged then you could get this result.  Perhaps your space group is really P31 or something like that?  What are your merging statistics? Try an analysis with data in lower-symmetry space groups  (you can always get the Wilson moments in P1 and that will often be a good indicator).

All the best,
Tom T

On Dec 10, 2014, at 1:37 PM, Xiao Lei wrote:

> Dear All,
>
> I have a x-ray dataset of a protein-DNA complex to 2.8 A resolution with space group P312 checked by phenix xtriage for twinning. The estimated twin fraction from the output of xtriage is: 0.115 (Britton analysis); 0.119 (H test) and 0.022 (maximum likelihood method). However, the L-test graph in xtriage shows my observed data almost perfectly overlay with theoretical perfect twin data. In addition, when I tried to use phenix to do refinement with twin law -h,-k,l, the log file shows my twin fraction estimation is 0.49, which is very high and much bigger than Britton analysis and H test estimation.
>
> As far as my understanding is that if a twin fraction is lower than 15%, I still have hope to solve the structure (molecular replacement in this case) with reasonable R value, but if the twin fraction is 0.49, which is almost a perfect twin, which makes detwin impossible and refinement will stall at high R values (in my case, R free start: 0.4199; R work start: 0.4121;  and R free final:  0.4038 and R work final: 0.3640 after running phenix refinement with twin law -h, -k, l).
>
> My question:  which twin fraction estimation is more reliable? is my data almost perfectly twined?
>
> I attached the graphs of L test, Britton analysis and twin estimation from phenix xtriage and part of log file from phenix refine here.
>
> Many thanks in advance.
>
> Xiao
> <p312_phenix_refine.png><P312_estimated twin_fraction.png><P312_Britton_plot.pdf><P312_Ltest.png>_______________________________________________
> phenixbb mailing list
> [email protected]
> http://phenix-online.org/mailman/listinfo/phenixbb