Hello Garib,
Thank you for the reply & helpful insight.
I ran the refinement of structure factor amplitudes without the TWIN command, and obtained much higher R and Rfree (32% and 36%). Also, in the process, I found out that I made a mistake earlier in that the amplitudes were refined as intensities.
To summarize the R and Rfree of the refinements:
Structure factors, no twin: 32%, 36%
Structure factors, twin: 21.5%, 25.0%
SF, twin, refined as intensities: 25%, 29% <-- wrong setting
Intensities, twin: 22.4%, 25.4%
These results fit into the general rule that you mentioned, and I believe I will proceed with the twinned refinement of structure factor amplitudes.
As you suggested, I also submitted the data+model to Zanuda on the server (see below), and it appears that the real space group is P6(5).
Best,
Peter
readability test passed (Refmac_5.5.0070)
resolution 1.947
spacegroup P 65
cell 96.530 96.530 91.930 90.00 90.00 120.00
Step 1.
R-factors for the starting model.
Transformation into a supergroup.
current time: Oct 13 22:21 BST
expected end of job (rough estimate): Oct 13 22:52 BST
---------------------------------------------------------------------
| Subgroup | Spacegroup | R.m.s.d. | Refinement in tested group |
| | | from the |--------------------------------|
| Ref | | starting | Rigid | Restrained |
| | | model, A |----------|---------------------|
| | | | R | R | R-free |
|----------|------------|----------|----------|----------|----------|
| >> 4 | P 65 | 0.0005 | -- | 0.2005 | 0.2230 |
| << 4 | P 65 | 0.0005 | -- | -- | -- |
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Step 2.
Refinements in subgroups.
There are 4 subgroups to test.
current time: Oct 13 22:32 BST
expected end of job: Oct 13 23:04 BST
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
| >> 4 | P 65 | 0.0005 | -- | -- | -- |
---------------------------------------------------------------------
| 1 | P 1 | 0.0711 | 0.2937 | 0.1949 | 0.2292 |
| 2 | P 1 21 1 | 0.0631 | 0.3217 | 0.1971 | 0.2305 |
| 3 | P 32 | 0.0629 | 0.3088 | 0.1938 | 0.2290 |
| 4 | P 65 | 0.0655 | 0.3321 | 0.1976 | 0.2316 |
---------------------------------------------------------------------
| << 3 | P 32 | 0.0629 | 0.3088 | 0.1938 | 0.2290 |
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Step 3.
Refinement of the best model.
Candidate symmetry elements are added one by one.
current time: Oct 13 22:43 BST
expected end of job: Oct 13 23:06 BST
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
| >> 3 | P 32 | 0.0629 | 0.3088 | 0.1938 | 0.2290 |
---------------------------------------------------------------------
| 1 | P 1 | 0.0714 | 0.2949 | 0.1952 | 0.2293 |
| 2 | P 1 21 1 | 0.0587 | -- | 0.1950 | 0.2306 |
| 4 | P 65 | 0.0596 | -- | 0.1961 | 0.2315 |
| 16 | P 65 2 2 | 1.6311 | -- | 0.4261 | 0.4499 |
---------------------------------------------------------------------
| << 4 | P 65 | 0.0596 | -- | 0.1961 | 0.2315 |
---------------------------------------------------------------------
R-factor in the original subgroup is (almost) the best.
The original spacegroup assignment seems to be correct.
end of job: Oct 13 23:02 BST
Date: Wed, 13 Oct 2010 21:51:57 +0100
From:
[log in to unmask]Subject: Re: [ccp4bb] Twin - Data reduction and refinement in Refmac
To:
[log in to unmask]As a general rule intensity based refinement gives higher R-factor especially when intensities are weak. Truncate smoothes out data and Rfactors become lower (it is not necessary that model becomes better)
In your case it may happen that your space group is higher. To check that you can do simple refinement and submit your data+model to Zanuda on the server:
It may be able correct your space group and give higer space group if it is the case.
What are Rfactors without twin refinement? If Rfactors with and without twin refinement are not very different then using twin would not make much sense.
regards
Garib
On 13 Oct 2010, at 16:01, Peter Chan wrote:
Hello All,
I am a graduate student working on my first merohedrally twinned data set. Like a few, I am a bit intimidated by it. After some trial and error with the help of some online resources and ccp4bb posts, I seem to have solved the structure. However, I am still unsure about some of the steps taken.
Background:
My dataset was processed to 1.95 A in XDS, the apparent space group is P622 (96 A, 96 A, 92 A, & 90, 90, 120). Various tests indicate that I have a twinned dataset. The Rsym of the data when processed in a lower symmetry space group (i.e. P6, P321 and P312) suggest that the real space group may be P6 because its Rsym is lower by ~1-2%. The screw axis could not be unambiguously identified from systematic absences. Molecular replacement by Phaser returned a solution in P6(5) with 2 molecules in the asymmetric unit. This was refined in Refmac with the twin option to R & Rfree of 22% and 25%. The twin fractions are 47% and 53%.
My questions and concerns:
- First and foremost, is there a chance that I may have processed the data in the wrong space group (or wrongly deduced the data in the right space group)?
- Should the diffraction data be merged or unmerged during the twin refinement in Refmac? The current dataset is fully merged (repeated measurements, Friedel pairs & symmetry related reflections). Would there be an improvement in the twin refinement if some of them are kept unmerged?
- Should the twin refinement be performed on the intensities or structure factor amplitudes? I have tried both (using the same set of Rfree flags): With intensities, the R/Rfree are 22%/25% and twin fractions are 47%/53%. With amplitudes, the R/Rfree are 25%/29% and twin fractions are 44.6%/55.4%. The resulting electron density maps are not significantly different, however. I don't understand why the statistics vary so significantly.
- During the model building, the electron density map appears to be 'weak'. Rebuilding some surface loops (by first deleting them, refining the omit structure, and remodeling into the difference map) which initially has some 2Fo-Fc density becomes tricky because there is not much density left in the refined omit structure. Was the initial densities purely a result of model bias or is this related to the twinning of the crystal? Could this also be because Refmac outputs a differently weighed 2Fo-Fc map (if I recall correctly) during twin refinement?
I would be very grateful for any help, comments and suggestions.
Best,
Peter Chan