Hi again,
Yes: provide a star file with 2 instances of rlnReferenceImage in a table
data_
loop_
_rlnReferenceImage
ref1.mrc
ref2.mrc
save this as refs.star, and indeed provide it with --ref refs.star --K 2
Be careful with the bias though.
HTH,
S
> Hi Sjors,
>
> Thanks for your reply. I meant we already knew that there are only two
> conformers, with/without ligand.
> Now I want to run this bipath 3D classification by providing two initial
> models (with/without ligand), how could I do this?
> Can I just input two models by '--ref' and '--K 2'?
>
> Thanks,
> Yan
>
>
>
> On Thu, Oct 23, 2014 at 12:33 PM, Sjors Scheres
> <[log in to unmask]>
> wrote:
>
>> Hi Yan,
>> I would be very careful with that. By providing your own references you
>> may bias the results towards something that does not happen in your
>> data...
>> S
>>
>> > Hi all,
>> >
>> > When we are talking about using the 3D classification to distinguish
>> the
>> > complex with/without ligand, could we introduce two different initial
>> > model
>> > (with/without ligand) as references to relion 3D classification?
>> >
>> > Thanks,
>> > Yan
>> >
>> > On Wed, Oct 22, 2014 at 2:28 PM, Sjors Scheres <
>> [log in to unmask]>
>> > wrote:
>> >
>> >> Hi Sebastian,
>> >>
>> >> That is a sharp observation. :-) I wouldn't worry too much about it
>> >> though. If classes remain similar, then particles will keep jumping
>> from
>> >> one class to the other. This is not necessarily a problem for
>> >> convergence.
>> >>
>> >> If the 3 classes that you see are similar, but you are not yet happy
>> >> with
>> >> the separation. You could try refining all particles from these 3
>> >> classes
>> >> against a single reference in the auto-refine tab. Then, input the
>> >> output
>> >> data.star file from that run as input into a 3D classification run.
>> In
>> >> that run, use a much finer angular sampling (e.g. as fine as the
>> >> estimated
>> >> accuracy in the auto-refine) and combine that with local angular
>> >> searches
>> >> in order to not to take too much CPU. We've had quite good results
>> with
>> >> this type of classifications in the past.
>> >>
>> >> Alternatively, you could design a (not too tight or too sharp) 3D
>> mask
>> >> around the region of the ligand, and provide this mask as the solvent
>> >> mask. Then, you could run a 3D classification where you skip the
>> >> alignment
>> >> of all particles. This may work well to separate with/without ligand,
>> >> but
>> >> be careful not to use too small a mask.
>> >>
>> >> HTH,
>> >> S
>> >>
>> >>
>> >>
>> >> > Hi,
>> >> >
>> >> > I have a question about 3D classification convergence. I want to do
>> a
>> >> 3D
>> >> > classification of a 4x decimated dataset (complex +/- ligand) of
>> about
>> >> > 80,000 particles into 4 classes. The particle diameter is 230 Å
>> with
>> >> a
>> >> > pixel size of 4.432 Å. First, I ran 20 iterations with 15°
>> sampling
>> >> and
>> >> > 1x oversampling until the classification converged, then I
>> increased
>> >> the
>> >> > sampling to 7.5°, again with 1x oversampling. However, now it
>> looks
>> >> like
>> >> > 20 - 25 % of all particles cannot be assigned to their optimal
>> >> classes.
>> >> >
>> >> > Now, I am a bit puzzled how to make sense of this. Does this mean,
>> >> that
>> >> > the classes are aleady to similar (3 of classes contain some kind
>> of
>> >> > ligand) or that I should use even more classes?
>> >> >
>> >> > I have attached the convergence plot to this email.
>> >> >
>> >> > Thank you very much for help,
>> >> >
>> >> > Sebastian Kube, PhD Student
>> >> > Group Dr. Petra Wendler
>> >> > Emmy-Noether Group Leader
>> >> > Ludwig-Maximilians-Universität Munich
>> >> > Gene Center Munich
>> >> > Feodor-Lynen-Straße 25
>> >> > 81337 Munich
>> >> >
>> >> > phone. +49 89 2180 76932
>> >> > email. [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
>> >>
>> >
>>
>>
>> --
>> 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
>>
>>
>
--
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
|