From: john peter
Date: 30 August 2011 20:31
Hello All,
This is regarding twinning in a data set.
I collected a native data set to resolution, 1.8 A. I used XDS suite
to process and scale the data set. It scaled well in P622 and I found
systematic absence (l=6n present).
Hence thought the space group may be P6122/P6522. SFCHECK did not
show any twinning and also it did not detect pseudo-translation. Twin
test in http://nihserver.mbi.ucla.edu/pystats/ ( Merohedral Twin
Detector: Padilla-Yeates Algorithm ) showed perfect twinning.
Scaled in P61/65 and SFCHECK reported twinning fraction 0.272 & no
pseudo-translation.
http://nihserver.mbi.ucla.edu/pystats/ ( Merohedral Twin Detector:
Padilla-Yeates Algorithm ) showed perfect twinning.
Another server from ucla http://nihserver.mbi.ucla.edu/Twinning/
showed partial twinning with twin fraction 0.23 as follows.
2 along a, b, a*, b*
No. Twin Law Related Reflections = 18033 (pairs)
No. Twin Law Pairs Considered = 9016
<H> = 0.266149
<H2> = 0.095303
Twin Fraction = 0.233249 +/- 0.000602
(SHELXL Commands: TWIN 1 0 0 -1 -1 0 0 0 -1 and BASF 0.233249)
In P61/65, I got the following matthews-coeffs
mol/asym Matthews Coeff %solvent P(1.73) P(tot)
_____________________________________________________________
1 9.75 87.39 .00 .00
2 4.88 74.79 .00 .01
3 3.25 62.18 .06 .14
4 2.44 49.57 .57 .63
5 1.95 36.97 .36 .21
6 1.63 24.36 .00 .00
7 1.39 11.75 .00 .00
_____________________________________________________________
May I ask what could be the real twin fraction and what is the
likelihood of solving the structure by molecular replacement by models
with 25 % sequence identity and 30 % sequence similarity.
Thank you so much for reading this mail during your busy hours and
all suggestions, comments would be gratefully welcome & appreciated.
thank you ccp4 mailing list.
John
----------
From: Garib N Murshudov
Space groups with point groups 622 and 432 merohedral twinning is not possible (they are the highest groups possible for proteins).
----------
From: <Herman.Schreuder
Dear John,
It is not the sequence identity/similarity that counts, but how similar
the protein folds are. For many protein families, the fold is identical
although the sequence identity is very low. With 25% sequence identity
and presumably a protein from the same family, I give you a good chance
that e.g. Phaser will find the solution. However, if there is a hinge
movement, or some other deformation, you may not find a solution despite
very high sequence identity.
Concerning your twinning: with 25% twinning fraction, you could try to
detwin using Yates algorithm, but I would just try Phaser first with the
data as is. I have done this with data sets with a twinning fraction
close to 50% and what happened was that phaser would find two solutions,
one for each twin orientation. The "raw" electron density map using this
twinned data was amazingly good, presumably because the data of the twin
fraction with the same orientation as the molecule used for phasing
would produce very nice electron density, while the data of the the twin
fraction with the other orientation would just produce noise, because
for this data phases would be more or less random.
So the bottom line: I would just run Phaser with the P6x data as is
(with SGALTERNATIVE ALL) and only if that does not work try more
elaborate methods.
Best,
Herman
----------
From: Eleanor Dodson
Dear John,
In a hexagonal crystal class the possible point groups are
P3 (only a 3-fold axis)
P321(3-fold plus 2-fold relating hkl and kh-l)
P312 (3-fold and 2-fold relating (hkl and -k,-h,-l)
P6 ( 6 fold axis only)
P622 (6-fold plus 2-fold relating hkl and k,h,-l)
pointless will tell you which of these symmetry operators is well observed, and suggest a pooint group.
truncate gives graphs of indicators of possible twinning.
The safest indicators are:
the moments of <E> - if the moment of <E>*4 is < 2.0 twinning is likely. However if you have non-crystallographic translation this can be misleading.
The L test done with the "safe" well-measured data
The cumulative intensity test with the "safe" well-measured data.
If these are satisfactory in your assignment of point group to P622 then that is probably the correct one.
If they are not then you need to choose which symmetry element to drop.
ctruncate will suggest which is the most likely twin operator, and the point group must be changed to remove that sym op (or sym ops)
The H test is not very reliable. it is only useful when you have the point group right..It looks at the agreement between symmetry pairs so obviously if you really have point group P6 but you assign the point group as P3, it will tell you that the symmetry pair h,k,l and -h,-k,l agree very well and give an H score of ~ 0.5.
Date: 30 August 2011 20:31
Hello All,
This is regarding twinning in a data set.
I collected a native data set to resolution, 1.8 A. I used XDS suite
to process and scale the data set. It scaled well in P622 and I found
systematic absence (l=6n present).
Hence thought the space group may be P6122/P6522. SFCHECK did not
show any twinning and also it did not detect pseudo-translation. Twin
test in http://nihserver.mbi.ucla.edu/pystats/ ( Merohedral Twin
Detector: Padilla-Yeates Algorithm ) showed perfect twinning.
Scaled in P61/65 and SFCHECK reported twinning fraction 0.272 & no
pseudo-translation.
http://nihserver.mbi.ucla.edu/pystats/ ( Merohedral Twin Detector:
Padilla-Yeates Algorithm ) showed perfect twinning.
Another server from ucla http://nihserver.mbi.ucla.edu/Twinning/
showed partial twinning with twin fraction 0.23 as follows.
2 along a, b, a*, b*
No. Twin Law Related Reflections = 18033 (pairs)
No. Twin Law Pairs Considered = 9016
<H> = 0.266149
<H2> = 0.095303
Twin Fraction = 0.233249 +/- 0.000602
(SHELXL Commands: TWIN 1 0 0 -1 -1 0 0 0 -1 and BASF 0.233249)
In P61/65, I got the following matthews-coeffs
mol/asym Matthews Coeff %solvent P(1.73) P(tot)
_____________________________________________________________
1 9.75 87.39 .00 .00
2 4.88 74.79 .00 .01
3 3.25 62.18 .06 .14
4 2.44 49.57 .57 .63
5 1.95 36.97 .36 .21
6 1.63 24.36 .00 .00
7 1.39 11.75 .00 .00
_____________________________________________________________
May I ask what could be the real twin fraction and what is the
likelihood of solving the structure by molecular replacement by models
with 25 % sequence identity and 30 % sequence similarity.
Thank you so much for reading this mail during your busy hours and
all suggestions, comments would be gratefully welcome & appreciated.
thank you ccp4 mailing list.
John
----------
From: Garib N Murshudov
Hello
If you could merge in 622 it means that Rmerge was very small. It is very likely that point group is either 622 or 6 with very strong rotational symmetry that is perpendicular to 6 fold axis. In these cases H test (sfcheck based on this) and N(z) test (truncate is based on this) will overestimate (H) or underestimate (N(z)) if twin is present. At the moment better test for these cases seems to be L-test (that is in ctruncate I think).
Solving structure using molecular replacement in the presence of twinning does not seem to be a problem (unless data are very noisy and/or model is too far).
Pointless gives very good idea about "true" space group.
I would solve in P6_{x} space groups and then run zanuda (from ysbl server) to correct space group.
I hope it helps
regards
Garib
Garib N Murshudov
Structural Studies Division
Structural Studies Division
MRC Laboratory of Molecular Biology
Hills Road
Cambridge
----------
From: <Herman.Schreuder
Dear John,
It is not the sequence identity/similarity that counts, but how similar
the protein folds are. For many protein families, the fold is identical
although the sequence identity is very low. With 25% sequence identity
and presumably a protein from the same family, I give you a good chance
that e.g. Phaser will find the solution. However, if there is a hinge
movement, or some other deformation, you may not find a solution despite
very high sequence identity.
Concerning your twinning: with 25% twinning fraction, you could try to
detwin using Yates algorithm, but I would just try Phaser first with the
data as is. I have done this with data sets with a twinning fraction
close to 50% and what happened was that phaser would find two solutions,
one for each twin orientation. The "raw" electron density map using this
twinned data was amazingly good, presumably because the data of the twin
fraction with the same orientation as the molecule used for phasing
would produce very nice electron density, while the data of the the twin
fraction with the other orientation would just produce noise, because
for this data phases would be more or less random.
So the bottom line: I would just run Phaser with the P6x data as is
(with SGALTERNATIVE ALL) and only if that does not work try more
elaborate methods.
Best,
Herman
----------
From: Eleanor Dodson
Dear John,
In a hexagonal crystal class the possible point groups are
P3 (only a 3-fold axis)
P321(3-fold plus 2-fold relating hkl and kh-l)
P312 (3-fold and 2-fold relating (hkl and -k,-h,-l)
P6 ( 6 fold axis only)
P622 (6-fold plus 2-fold relating hkl and k,h,-l)
pointless will tell you which of these symmetry operators is well observed, and suggest a pooint group.
truncate gives graphs of indicators of possible twinning.
The safest indicators are:
the moments of <E> - if the moment of <E>*4 is < 2.0 twinning is likely. However if you have non-crystallographic translation this can be misleading.
The L test done with the "safe" well-measured data
The cumulative intensity test with the "safe" well-measured data.
If these are satisfactory in your assignment of point group to P622 then that is probably the correct one.
If they are not then you need to choose which symmetry element to drop.
ctruncate will suggest which is the most likely twin operator, and the point group must be changed to remove that sym op (or sym ops)
The H test is not very reliable. it is only useful when you have the point group right..It looks at the agreement between symmetry pairs so obviously if you really have point group P6 but you assign the point group as P3, it will tell you that the symmetry pair h,k,l and -h,-k,l agree very well and give an H score of ~ 0.5.
No comments:
Post a Comment