Dear Grigory,
rlnCoarseImageSize and rlnCurrentImageSize are internal parameters over
which you have no control. They reflect the current resolution of the
refinement. If they stay low, your resolution is just not progressing
much, but that has not necessarily anything to do with the pixel size. So,
if the refinement converges in similar ways as before, I would not worry
about it. Using smaller pixel sizes does not necessarily increase
resolution. Because RELION interpolates very well almost up to the Nyquist
frequency, usually only when the FSC stays above 0.143 with the
downsamples data, it is useful to go back to the original images.
HTH,
S
> Dear Sjors,
>
> I have managed to continue from some intermediate iteration and changing
> only rlnDetectorPixelSize, rlnOriginX, rlnOriginY,
> rlnMaximumCoarseImageSize, rlnOriginalImageSize and rlnPixelSize. But even
> if the refinement has converged, I can see in optimiser.star and
> model.star
> files that rlnCoarseImageSize and rlnCurrentImageSize didn't even reach
> half size on the new box (except last iteration, where full box is used).
>
> My question is should I multiply by 2 also rlnCoarseImageSize and
> rlnCurrentImageSize
> when I change coarse level and continue local searches?
>
> Best regards,
> Grigory Sharov
>
> Institute of Genetics and Molecular and Cellular Biology
> Department of Structural Biology and Genomics
> 1, rue Laurent Fries
> 67404 Illkirch, France
> tel. 03 69 48 51 00
> e-mail: [log in to unmask]
>
>
> On Mon, Feb 16, 2015 at 2:05 PM, Grigory Sharov <[log in to unmask]>
> wrote:
>
>> Dear Sjors,
>>
>> I have done autorefine with particles coarsened by factor of 4, it has
>> reached Nyquist, and I'd like to continue with particles coarsened by
>> factor of 2. I have re-extracted particles (with relion_preprocess) and
>> rescaled models / half models / mask to correct pixel and box size (with
>> relion_image_handler). I have also changed the following fields in star
>> files:
>>
>> *data.star*: in data_images table
>> _rlnImageName
>> _rlnDetectorPixelSize (divide by 2)
>> _rlnOriginX (multiply by 2)
>> _rlnOriginY (multiply by 2)
>>
>> *optimiser.star*: in data_optimiser_general table
>> _rlnOutputRootName
>> _rlnModelStarFile
>> _rlnModelStarFile2
>> _rlnExperimentalDataStarFile
>> _rlnOrientSamplingStarFile
>> ...
>> _rlnCoarseImageSize (multiply by 2)
>> _rlnMaximumCoarseImageSize (multiply by 2)
>>
>> *half1(2)_model.star*: in data_model_general table
>> _rlnOriginalImageSize (multiply by 2)
>> _rlnCurrentImageSize (multiply by 2)
>> _rlnPixelSize (divide by 2)
>>
>> *half1(2)_model.star*: in data_model_classes table
>> _rlnReferenceImage
>>
>>
>> But after launching autorefine in continue mode, I get almost
>> immediately
>> an error:
>>
>> Auto-refine: Iteration= 16
>>> Auto-refine: Resolution= 16.1143 (no gain for 4 iter)
>>> Auto-refine: Changes in angles= 3.53268 degrees; and in offsets=
>>> 0.700853 pixels (no gain for 5 iter)
>>> Estimating accuracies in the orientational assignment ...
>>> 1/ 1 sec
>>> ............................................................~~(,_,">
>>> Auto-refine: Estimated accuracy angles= 0.05 degrees; offsets= 0.05
>>> pixels
>>> Auto-refine: Angular step= 1.875 degrees; local searches= true
>>> Auto-refine: Offset search range= 1.75213 pixels; offset step=
>>> 0.219017
>>> pixels
>>> CurrentResolution= 16.1143 Angstroms, which requires
>>> orientationSampling
>>> of at least 4.28571 degrees for a particle of diameter 430 Angstroms
>>> Oversampling= 0 NrHiddenVariableSamplingPoints= 14450688
>>> OrientationalSampling= 3.75 NrOrientations= 140
>>> TranslationalSampling= 0.438033 NrTranslations= 49
>>> =============================
>>> Oversampling= 1 NrHiddenVariableSamplingPoints= 462422016
>>> OrientationalSampling= 1.875 NrOrientations= 1120
>>> TranslationalSampling= 0.219017 NrTranslations= 196
>>> =============================
>>> Estimated memory for expectation step > 0.240457 Gb, available memory
>>> =
>>> 32 Gb.
>>> Estimated memory for maximization step > 0.249878 Gb, available memory
>>> =
>>> 32 Gb.
>>> Expectation iteration 16
>>> 000/??? sec ~~(,_,">
>>> [oo] exp_thisparticle_sumweight= exp_thisparticle_sumweight= 0
>>> 0
>>> exp_thisparticle_sumweight= 0
>>> exp_part_id= 13438exp_iimage=1
>>> group_id= 128 mymodel.scale_correction[group_id]= 1.0384
>>> exp_ipass= 1
>>> sampling.NrDirections(0, true)= 3072 sampling.NrDirections(0, false)=
>>> 30
>>> sampling.NrPsiSamplings(0, true)= 96 sampling.NrPsiSamplings(0,
>>> false)= 6
>>> mymodel.sigma2_noise[exp_ipart]=
>>
>> 0
>>> 0
>>> 0
>>> 0
>>> 0
>>>
>> .......................
>>
>> 0
>>> 0
>>> mymodel.avg_norm_correction= 0.845458
>>> wsum_model.avg_norm_correction= 0
>>> written out Mweight.spi
>>> exp_thisparticle_sumweight= 0
>>> exp_min_diff2[exp_ipart]= 9.9e+100
>>> exp_thisparticle_sumweight= 0
>>> slave 2 encountered error: ERROR!!! zero sum of weights....
>>> File: src/ml_optimiser.cpp line: 3982
>>
>>
>> I didn't find any black or white particles in re-extracted dataset, I
>> also
>> tried to switch off normalisation and/or scale correction in optimiser
>> file, regroup particles, but always get the same error.
>>
>>
>> Best regards,
>> Grigory Sharov
>>
>> Institute of Genetics and Molecular and Cellular Biology
>> Department of Structural Biology and Genomics
>> 1, rue Laurent Fries
>> 67404 Illkirch, France
>> tel. 03 69 48 51 00
>> e-mail: [log in to unmask]
>>
>>
>
--
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
|