Dear Ambroise,
The problem you describe only occurs at Class2D in the current version. in
Refine3D, translations are only searched -1/2 rise to +1/2rise, as
described in the helical reconstruction paper. Therefore, we often
re-extract particles, while resetting their translational shifts after
initial rounds of Class2D and before enetering Refine3D. In the upcoming
version of relion, there will be an option for Class2D to also limit
translational searches along the helicl axis to 1 rise.
So, for now: reset your translations after Class2D, and then Refine3D will
be fine.
HTH,
Sjors
> Hi Takanori,
>
> thanks for your answer.
>
> Here we refer only to Refine3D (not Class2D), straight after extraction.
> I can see the priors _rlnAngleTiltPrior and _rlnAnglePsiPrior, as well
> as the helical tracklength _rlnHelicalTrackLength. When you ask whether
> we "change such priors to increase the search range", do you refer to
> --offset_range, or to another prior ? We used an offset range that
> correspond to about half the helical rise and takes into account
> possible mis-centering of the segments. With the offset priors,
> relatively high translations can still accumulate over the iterations,
> and even with very small offsets, the problem that I describe would
> persist for filaments with small helical rise and/or low
> --helical_nr_asu used for extraction.
>
> Any segment orientation could be brought -between iterations- to the
> lowest translation possible along the helical axis, by applying a
> multiple of the helical rise as shift along helix axis and the
> corresponding change in _rlnAngleRot by the same multiple of the helical
> twist. This would ensure no such "duplicates". But this is not what you
> refer to as "prior on translation", is it ? In a case with one segment
> extracted per asymmetric unit, subsequent segments would then have their
> relative _rlnAngleRot spaced by about the helical twist. This is what we
> observe for a good percentage of the segments, but far from all : many
> subsequent segments have similar _rlnAngleRot or are spaced by twice the
> _rlnAngleRot.
>
> With best regards,
>
> Ambroise
>
> On 5/22/19 3:05 PM, Takanori Nakane wrote:
>> Hi,
>>
>> In Refine3D, RELION applies a prior on translation search to prevent
>> the problem you described.
>>
>> In Class2D of RELION 3.0 or earlier, such translation prior is not
>> applied.
>> So you have to be careful when re-extracting particles after Class2D.
>> "Reset the refined offsets to zero?" is useful in such cases.
>> See the help message:
>> "If set to Yes, the input origin offsets will be reset to zero.
>> This may be useful after 2D classification of helical segments,
>> where one does not want neighbouring segments to be translated on
>> top of each other for a subsequent 3D refinement or classification."
>>
>> In RELION 3.1, Class2D can also use the helical translation prior.
>>
>>> one has to give enough freedom for the offset search range to correctly
>>> center the segments and align them to the projection of the reference.
>> Did you change such priors to increase the search range?
>>
>> Best regards,
>>
>> Takanori Nakane
>>
>>> Dear all,
>>>
>>> we noticed an issue with alignment parameters from helical
>>> reconstruction in Relion (but it is not specific to Relion), and would
>>> like to know if someone has the script that fixes this issue -or if we
>>> should write something and share it later.
>>>
>>> When working with helical reconstruction, input segments have generally
>>> a large overlap. In the most extreme case, there is one segment per new
>>> asymmetric unit, and the distance between segments is equal to the
>>> helical rise. During 3D refinement (here auto-refine from Relion3), one
>>> has to give enough freedom for the offset search range to correctly
>>> center the segments and align them to the projection of the reference.
>>> The result of this is that 2 (or more) subsequent segments can align
>>> with the same euler angles (or close), just by shifting more along
>>> helical axis. Therefore, some subunits are included multiple times at
>>> same 3D position (which increases noise), and some subunits are not
>>> included at all (which decreases signal).
>>>
>>> We are aware that a general solution to this problem is not easy to
>>> implement during refinement (it depends on helical symmetry, expected
>>> centering offset, distance between segments, etc), but we were
>>> wondering
>>> if someone has a script to convert the output alignment parameters from
>>> a refinement by taking into account the helical symmetry and these
>>> "duplicated" segments, to create a new alignment parameter file by
>>> shifting offsets and correcting euler angles of duplicated segments.
>>> Such corrected alignment parameter file could then be used for a few
>>> more refine iterations with local searches of angles and restricted
>>> shifts.
>>>
>>> In one of our datasets, although the distance between successive
>>> segments is rather large (~20 pixels), this problem is strongly
>>> present.
>>>
>>> In addition to the loss of information, this problems prevent us for
>>> doing a proper particle subtraction step (since two successive
>>> subtracted images then contain the exact same subunit..).
>>>
>>> Thanks much for any help
>>>
>>> Ambroise
>>>
>>> ########################################################################
>>>
>>> To unsubscribe from the CCPEM list, click the following link:
>>> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
>>>
>> ########################################################################
>>
>> To unsubscribe from the CCPEM list, click the following link:
>> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
>
> ########################################################################
>
> To unsubscribe from the CCPEM list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
>
--
Sjors Scheres
MRC Laboratory of Molecular Biology
Francis Crick Avenue, Cambridge Biomedical Campus
Cambridge CB2 0QH, U.K.
tel: +44 (0)1223 267061
http://www2.mrc-lmb.cam.ac.uk/groups/scheres
########################################################################
To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
|