First the stats for POINTLESS with XDS_ASCII.HKL (CORRECT with SPACE_GROUP_NUMBER= 3
):
Nelmt Lklhd Z-cc CC N Rmeas Symmetry & operator (in Lattice Cell)
1 0.913 9.30 0.93 50862 0.039 identity
2 0.915 9.23 0.92 93951 0.045 *** 2-fold l ( 0 0 1) {-h,-k,l}
3 0.916 9.21 0.92 93677 0.050 *** 2-fold k ( 0 1 0) {-h,k,-l}
4 0.915 9.24 0.92 96828 0.045 *** 2-fold h ( 1 0 0) {h,-k,-l}
Interestingly, CORRECT.LP with SPACE_GROUP_NUMBER= 0
and unspecified TEST_RESOLUTION_RANGE
:
SPACE-GROUP UNIT CELL CONSTANTS UNIQUE Rmeas COMPARED LATTICE-
NUMBER a b c alpha beta gamma CHARACTER
5 72.0 37.7 126.2 90.0 90.1 90.0 1372 12.0 3617 39 mC
5 37.7 72.0 126.2 90.0 90.0 90.0 1325 13.0 3664 29 mC
1 37.7 40.6 126.2 89.9 90.0 62.4 2499 8.6 2490 31 aP
21 37.7 72.0 126.2 90.0 90.0 90.0 738 18.4 4251 38 oC
* 3 37.7 126.2 40.7 90.0 117.7 90.0 1303 12.6 3686 34 mP
1 37.7 40.6 126.2 90.1 90.0 117.6 2499 8.6 2490 44 aP
Why did xds prefer “3” over “5”, which has a lower Rmeas?
Jacob asked about the refmac-assigned twin ratio:
Twin operator: H, K, L : Fraction = 0.519; Equivalent operators: -H, K, -L
Twin operator: -H, -K, H+L: Fraction = 0.481; Equivalent operators: H, -K, -H-L
Is it reasonable to compare these values with the L-test ratio? It may be relevant that the crystal includes a significant (> 50% of macromolecular non-H atoms) DNA double helix component. Could the DNA exert a translational pseudo-symmetry effect on the intensity ratios and mask a truly higher twin ratio?
On Fri, 8 Jul 2016 17:14:42 -0400, wtempel <[log in to unmask]> wrote:
>Hello all,
>expanding this thread, and keeping Garib’s paper in mind, how would my
>colleagues proceed in the following case:
>1.8 Å data can can be merged in space group C2221 with an Rmeas of 5%. The
it pays off to look very closely how well it can be merged. E.g. only if the twin fraction alpha is high (say, >0.4) you get the same good Rmeas in the apparent (wrong) high-symmetry space group as in the correct low-symmetry space group. For your low twin fraction, I would expect that merging in P21 gives a better Rmeas than in C2221. In case you use XDS, maybe you want to use higher resolution data to do that comparison - adjust TEST_RESOLUTION_RANGE ! Pointless automatically finds the appropriate resolution range for the comparison, and gives very detailed information about how well the data support each symmetry element. So please post here what you get - for a true orthorhombic crystal you should get something like
*******************************************
Analysing rotational symmetry in lattice group P m m m
----------------------------------------------
<!--SUMMARY_BEGIN-->
Scores for each symmetry element
Nelmt Lklhd Z-cc CC N Rmeas Symmetry & operator (in Lattice Cell)
1 0.941 9.62 0.96 175778 0.054 identity
2 0.942 9.61 0.96 227697 0.053 *** 2-fold l ( 0 0 1) {-h,-k,l}
3 0.942 9.61 0.96 225738 0.053 *** 2-fold k ( 0 1 0) {-h,k,-l}
4 0.942 9.61 0.96 225139 0.053 *** 2-fold h ( 1 0 0) {h,-k,-l}
In this case all symmetry operators are as good as the identity operator, so it is true crystallographic symmetry (or the twinning is perfect, but this is what the L-test would tell you).
HTH,
Kay
>L-test suggests a twin ratio of 0.13 and Rwork/Rfree hover around 0.44/0.48
>with an essentially complete structure. After expansion to P21, and
>twin-aware assignment of free flags, Rwork/Rfree are 0.24/0.29 without and
>0.19/0.26 with twin refinement in refmac. The coordinates do not differ in
>any obvious way between runs with or without twin refinement. Is this
>sufficient evidence to rule out C2221? If so, how would readers of this
>discussion forum decide which, if any, should become the “model of record”:
>refined with or without the twinning option.
>With best regards,
>Wolfram Tempel
>
>
>On Wed, Jul 6, 2016 at 9:05 AM, Kay Diederichs <
>[log in to unmask]> wrote:
>
>> On Wed, 6 Jul 2016 09:13:22 +0100, Randy Read <[log in to unmask]> wrote:
>>
>> >Dear Zbyszek,
>> >
>> >I agree completely with your general point that there is a trend for an
>> increasing number of people to adopt too-low symmetry, rewarded by lower
>> R-factors in twinned refinement.
>>
>> Dear Randy,
>>
>> I wish you had not used the word "rewarded" - for Germans at least, it has
>> no ironic or pejorative connotation. I hope people do not understand this
>> as if you were endorsing this unfortunate practice. There are already too
>> many structures being twin-refined "because it reduces the R-factors" (and
>> I fell into this trap as well before reading Garib N. Murshudov (2011)
>> "Some properties of crystallographic reliability index - Rfactor: effect of
>> twinning" Appl. Comput. Math., V.10, N.2, 2011, pp.250-261
>> http://www.ysbl.york.ac.uk/refmac/papers/Rfactor.pdf ).
>>
>> best,
>>
>> Kay
>>
>