[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [ccp4bb]: perfect twinning
*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
hello!
first of all: this will sound a little disappointing.... sorry
from my experience with replacement of perfectly twinned crystals (so far two
proteins), it is possible to get a correct replacement-solution by following the
stategy described in "THE twin-paper":
first do a rotation-search in the apparent space group, then do the translation
search in the real space group (can be conveniently done with molrep). the
advantage is that you won't get overlapping molecules as you already decide for
only one of the twins in your rotation search.
ok. so far it sounds promissing. r-values decreased to slightly over 30 - and
stuck (refinement done with cns-twin-refine; that means you apply the twin-law to
your f-calc and compare these to your f-obs. you should fix your twin ratio for
this even with not perfectly twinned data, otherwise - for not really clear
reasons - you can refine everything!) the problem starts now: even though
additional density shows up omit-maps really look disencouraging e.g. you omit
conserved parts from your model, where it shows good density and it won't show
up in the newly calculated map...
I calculated maps with the cns-script. to calculate maps for a perfect twin, you
MUST have a model.
best aproximation is:
with TWOP beeing the twin-operator in the "hkl-form"
and Icalc('hkl') is the calulated twinned intensity (the 0.5 are not really
important... )
as
Iobs('hkl') = 0.5 I(hkl) + 0.5 I(TWOPhkl)
Icalc('hkl') = 0.5 Icalc(hkl) + 0.5 Icalc(TWOP[hkl])
so
FRAC(hkl) = Icalc(hkl)/Icalc('hkl')
that means if you aply FRAC(hkl) to your Iobs('hkl') you should get the best
aproximation with regard to your Iobs(hkl)..... (BEWARE: FRAC has NOTHING to do
with your twin-fraction! it is the fraction of the 'real' reflex hkl contributing
to the 'twinned' reflection 'hkl'; it is a number different for EVERY reflection
you have!) however as this fraction is to be computed for each reflexion and is
solely dependend on your actual model you end up with nearly no new information
:-((
another try was - as I had several molecules asymmetric - a 'cyclic' detwinning.
this should reduce model-bias. the idea was:
0. compute Fcalc and phicalc with model in the real space-group
determine ncs-ops
determine monomer masks
1. take Fcalc
2. square them
3. apply twin-law
4. determine fraction for each reflection
5. calculate your theoretical I(hkl) from I('hkl')
6. truncate (will give detwinned "Fobs")
7. phase with phicalc from step 1. (for a "2fofc" you use fc from step 1 but in
this case a "Fobs"-map should work better)
8. average density with ncs-op and mask from 0.
9. back transform density
10. goto 1.
calulations can be done with sftools... averaging can be performed with main, as
you can conveniently create masks and operators and transform forwards and
backwards.. and you see what is happening ;-)
while the idea was, that the estimate of the fraction of I(hkl) and I(TWOP[hkl])
gradually improves and 'forgets' about the model, I ended up with somthing that
could be displayed on the screen but had nothing to to with an electron density
of a protein ... I didn't try on, however for the desperate it would be worth
another try....
but all in all i think that efforts in recrystallizing and/or recloning (other
constructs, a little shorter/longer, fusion-proteins etc.) are worth more energy
than trying to refine a perfect twin.... (by the way: I was told this also and
didn't belive it first.... ;-) )
but probably the success is strongly dependent on the space-group, data-quality
and 'amount' of data (i.e. resolution....)
good luck anyway!
jens
raji wrote:
> *** For details on how to be removed from this list visit the ***
> *** CCP4 home page http://www.ccp4.ac.uk ***
>
> Yes, CNS does have a routine to detwin perfect hemihedral twins
> (detwin_perfect.inp).
> It detwins data based on a twinning operator, twinning fraction and model
> amplitudes. I quite dont understand how that works though but I think it works
> better when you have a starting model.
> Raji
>
> >===== Original Message From Bjoern Mamat
> <Bjoern.Mamat@mpibp-frankfurt.mpg.de> =====
> >*** For details on how to be removed from this list visit the ***
> >*** CCP4 home page http://www.ccp4.ac.uk ***
> >
> >>From my very limited experience, I got the impression that
> >CNS can handle partial, but not perfect merohedral twinning.
> >
> >To my knowledge, all structures solved using perfect twins so far
> >have been refined with SHELXL --
> >
> >any comments on that?
> >
> >-- bjoern
> >
> >>
> >> *** For details on how to be removed from this list visit the ***
> >> *** CCP4 home page http://www.ccp4.ac.uk ***
> >>
> >> > Li Minghui wrote:
> >> >
> >> > Dear all,
> >> >
> >> > I have a perfectly twinned dataset.
> >> >
> >> > At first, I determined that the spacegroup was P43212 and could find
> >> > MR solution using Molrep. There were two molecules in the ASU, and the
> >> > initial R-factor was about 0.5. After a cycle of rebuilding and
> >> > refinement the R-factor dropped to 0.36. The density map calculated
> >> > from the new model was good, and some omitted parts of the initial
> >> > model also have clear densities. I found there were also many other
> >> > densities outside the molecules. I could dock part of an other model
> >> > in them, but the remaining parts would overlap on the existing
> >> > molecules.
> >> >
> >> > I processed the data again in P4 spacegroup and found the data was
> >> > perfectly twinned with a twinning fraction of 0.48. It is said that
> >> > perfect twinned data is difficult to detwin.
> >> >
> >> > Does anybody have successful experience with perfect twinned data? Any
> >> > suggestion will be apperciated.
> >> >
> >> > Minghui Li
> >> >
> >> > National Laboratory of Biomacromolecules
> >> > Institute of Biophysics, Academia Sinica
> >> > Tel: 86-10-64888507
> >> > E-mail: lmh@moon.ibp.ac.cn
> >>
> >> You can indeed solve a MR search against twinned data - you will find
> >> two overlapping molecules of course..
> >>
> >> SHELX and CNS can both refine your models against the twinned data, and
> >> also refine the twinning factor.
> >>
> >> You will have difficulties with generating maps for rebuilding though..
> >> There are ways to detwin the Iobs using the calculated values as a guide
> >> but I have no experience of how much bias it introduces.
> >>
> >> Eleanor
> >
> >--
> >----------------------------------------------------------------------------
> >Bjoern Mamat mamat@mpibp-frankfurt.mpg.de
> >----------------------------------------------------------------------------
> >Max-Planck-Institute of Biophysics http://www.biophys.mpg.de
> >Heinrich-Hoffmann-Str. 7 phone +49 69 96769 384/393
> >60528 Frankfurt/Main, Germany fax +49 69 96769 423
> >----------------------------------------------------------------------------
>
> Rajeswari Edayathumangalam
> Dept. of Biochemistry & Molecular Biology
> Colorado State University
> Fort Collins, CO 80523 USA
--
--------------------------------------------------------------------------------
| Dr. Jens T. Kaiser | Phone: +49-89-8578-2732 |
| Max-Planck of Biochemistry | FAX: +49-89-8578-3516 |
| Am Klopferspitz 18a | Private: +49-89-7245-9547 |
| D-82152 Martinsried | Mobile: +49-160-990-85220 |
| Germany | e-mail: kaiser@biochem.mpg.de |
--------------------------------------------------------------------------------