Dear Kyle,
I'm not sure. The number of sampled orientations is also important. As
said before: the actual values don't matter much.
Best,
Sjors
> Hi Sjors,
>
> Thanks! I’ll take a look.
>
> To come back to symmetry and MaxValueProbabilityDistribution (MVPD), for
> another particle set I’m working on right now the distribution of these
> scores in general increase when not imposing symmetry (see attached). If I
> understood, for a symmetrical particle and refining without symmetry the
> angular assignments would be less certain when searching the whole of
> angular space due to multiple asymmetric unit views being found.
>
> Can you suggest why I see the opposite trend here, where angular
> assignment accuracy as indicated by the MVPD decrease when searching only
> the asymmetric unit having imposed symmetry?
>
> Thanks again for your help,
> Kyle
>
> [cid:E3482760-7D5B-4ED1-A8F2-EF50F505B970@warwick.ac.uk]
>
> On 12 Dec 2015, at 10:24, Sjors Scheres
> <[log in to unmask]<mailto:[log in to unmask]>> wrote:
>
> Hi Kyle,
> Sure. For each group, the sigam_noise estimates are stored in the
> model.star files in tables called data_model_group_X (X being the group
> number). In the data_model_groups table you can see the names of all
> groups as well.
> HTH,
> Sjors
>
> Hi both,
>
> Thanks for your contributions and explanation here.
>
> Sjors, I’m curious then, is it possible then to extract Relion’s
> sigma-noise estimates? To check they are in fact different, follow the
> same trend and thus are responsible for the differences in my
> LogLikeliContribution.
>
> Thanks for you clarification on the way symmetry is handled and the
> probability distribution.
>
> Best,
> Kyle
>
> On 10 Dec 2015, at 21:11, Sjors Scheres
> <[log in to unmask]<mailto:[log in to unmask]>>
> wrote:
>
> Hi Kyle,
> The _rlnLogLikeliContribution is higher for better particles. Just like
> the cross-correlation coefficient it isn't a very good discriminator
> though. It is also very much depending on sigma-noise variables, which
> may
> differ throughout your data set. We usually don't throw away particles
> based on this.
>
> The _rlnMaxValueProbabilityDistribution has been reported by Bridget
> Carragher (20S proteasome paper in eLife) to be useful for selecting
> particles. Higher values (closer to 1) mean sharper posterior
> probability
> distributions for the orientations, and possibly thus better particles.
> But in many cases, in our experience this is not a very good
> discriminator. For symmetric particles only the 'asymmetric unit' of
> the
> rotational space is searches, so symmetry doesn't matter.
>
> In our experience the best discriminator for good/bad particles is
> 2D/3D
> classification.
>
> HTH,
> Sjors
>
>
>
> Dear Sjors and CCPEM community,
>
> Can I seek some clarification on the _rlnMaxValueProDistribution and
> _rlnLogLikeliContribution please?
>
> On the wiki it says _rlnLogLikeliContribution is equivalent to the
> phase
> residual or cross correlation metric of other refinement programs. In
> the
> former case you want low values but in the latter case high values if
> I’m not mistaken. In trying to understand which particles to remove:
>
> 1) What should we expect for good agreement of particles to model in
> Relion’s convention for _rlnLogLikeliContribution?
>
> You can see in the attached that there are two clear groups in the
> _rlnLogLikeliContribution per particle. This same data set also gives,
> what I presume are, very low _rlnMaxValueProDistribution (MVPD) for
> the
> majority of the particles, see attached, particle # on x, MVPD on y
> but
> there is no correlation between these two metrics.
>
> 2) Should one expect a correlation between _rlnLogLikeliContribution
> and
> _rlnMaxValueProbabilityDistribution?
>
> I wonder if it is possible that one would expect a low
> _rlnMaxValueProbabilityDistribution for a symmetrical particle where
> different symmetrically related angular assignment can each match your
> model. In this case the reported estimated angular accuracy is < 1
> degree.
>
> 3) Is there consideration in Relion-1.4 to treat the calculation of
> _rlnMaxValueProbabilityDistribution differently based on having
> symmetry
> and if not would we expect angular some uncertainty due to having a
> symmetrical particle? i.e. Is what we see unusual?
>
> Thanks for your help!
>
> Kyle
>
> [cid:DE7B6796-12FD-44D2-9F79-7765B0AF407F@warwick.ac.uk]
>
> ,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-
> ‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-
> Kyle L. Morris BSc PhD
> Post-doctoral Research Fellow
> School of Life Sciences
> University of Warwick
> UK
>
> [log in to unmask]<mailto:[log in to unmask]><mailto:[log in to unmask]>
> https://sites.google.com/site/kylelmorris/
> ,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-,-
> ‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-‘-
>
>
>
>
> --
> 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
|