Hi Mani,
I'm afraid I have not much more to add than what is in the tutorial/Wiki
and what we discussed before. Given Hongwei's comments, it might be that
your data are just not good enough for 3D classification.
Best,
S
On 12/16/2014 11:18 AM, Manikandan KARUPPASAMY wrote:
> Hi Sjors,
>
> As recommanded, now i have tried the 3d classification with single
> reference. The output volumes are very similar after 25 iterations.
> The particle distributions are 1-2% different between them.
>
> Actually, from the RCT reconstructions, we have different shaped
> volumes (globular, L-shape and curved L-shape). Therefore the
> consensus volume from the
> entire dataset does not represent any of these.
>
> Should I have to increase the number of iterations ?
>
> Anyother parameter needs to be played with in 3d classification?
>
> thanking you
>
> with kind regards
>
> Mani.
>
> On 2014-12-14 11:10, Sjors Scheres wrote:
>> Hi Mani,
>> The recommended way is to use a single reference. Supervising the
>> classification like you did may sometimes work, but has it's dangers.
>> There is no auto-refine-like procedure in classification that changes
>> the
>> sampling rate, but this is not likely your problem. Classifying stain
>> data
>> is usually more difficult than cryo-EM data. Stain is different in many
>> different particles, and other artefacts may arise. Also, it might be
>> that
>> your RCT models are not good enough.
>> HTH,
>> S
>>
>>> Hi Relion Users,
>>>
>>> I have been trying to classify a heterogeneous negative stain data. The
>>> starting references (5 different volumes) are from RCT reconstructions.
>>> The input dataset is 70,000 particles.
>>>
>>> mpirun -np 108 -machinefile cluster.txt `which relion_refine_mpi` --o
>>> Class3D/r3_k5_7p5d --i all_untilt.star --particle_diameter 320 --angpix
>>> 5.71 --ref ref_volumes.star --firstiter_cc --ini_high 60 --iter 25
>>> --tau2_fudge 3 --K 5 --flatten_solvent --zero_mask --strict_highres_exp
>>> 10 --oversampling 1 --healpix_order 2 --offset_range 5 --offset_step 2
>>> --sym C1 --norm --scale --j 1 --memory_per_thread 4
>>>
>>>
>>> The output volumes are much distorted as compared to initial ones and
>>> the particle distributions are almost equal between the 3d classes.
>>> Moreover, when i compared the 3d projections of some of the refined
>>> volumes (auto_refine on the classified sub-pool of particles) against
>>> the reference free 2d class averages, the projections show more
>>> features than actual 2d class averages.
>>>
>>> Is there a better sort out the heterogeneous data set in relion?
>>> (classification based on the consensus reference model did not work
>>> also).
>>>
>>> Is it true that 3d classification runs only on the selected angular
>>> sampling?
>>>
>>> Is it possible make it like in '3d_auto_refine' where angular sampling
>>> is calculated by the program for every iteration?
>>>
>>> thanking you.
>>>
>>> with kind regards
>>>
>>> Mani.
>>>
>
--
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
|