Dear Thomas,
the reason for the collapse of all particles in a single class is the
attraction problem described in Eq. 6 ofth C.O.S. Sorzano, J.R.
Bilbao-Castro, Y. Shkolnisky, M. Alcorlo, R. Melero, G.
Caffarena-Fernandez, M. Li, G. Xu, R. Marabini, J.M. Carazo. A
clustering approach to multireference alignment of single-particle
projections in electron microscopy. J. Structural Biology 171: 197-206
(2010). Basically, when the algorithm has to decide a class for an
image, the absence of noise in the class with most particles puts an
important barrier to the remaining class averages because the
contribution of the clean background to the least squares is smaller.
This behavior (maybe not so pronounced) is rather usual in most
classification algorithms unless they are designed with specific
countermeasures. Xmipp CL2D and GL2D, and Eman bispectrum (all of them
available through Scipion) are much less sensitive to this effect
(probably others too, but I don't have so much experience with them).
Kind regards, Carlos Oscar
El 06/09/2019 a las 13:27, Thomas Bausewein escribió:
> Dear all,
>
> after doing particle polishing, which looks pretty normal, I did a classification in 2D (30 classes with ignore first peak option) or in 3D (4 classes, full ctf) with the polished particles.
> The result is, that all particles end up in a single class with about 95 % of the particles while the other classes are basically empty.
> This happens especially in 3D, while in 2D this effect is only seen in the first 15 iterations.
> However, the single class looks good, but the other classes are somewhat grainy.
>
> Without polishing this does not happen and the distribution of particles is rather even.
>
> Has anyone else experienced this effect?
> Or does anyone know what could be the reason for it?
>
> I have seen this behavior before, even with different proteins / samples / datasets.
>
> Thank you very much,
> Thomas
>
> ########################################################################
>
> To unsubscribe from the CCPEM list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
--
------------------------------------------------------------------------
Carlos Oscar Sánchez Sorzano e-mail: [log in to unmask]
Biocomputing unit http://i2pc.es/coss
National Center of Biotechnology (CSIC)
c/Darwin, 3
Campus Universidad Autónoma (Cantoblanco) Tlf: 34-91-585 4510
28049 MADRID (SPAIN) Fax: 34-91-585 4506
------------------------------------------------------------------------
########################################################################
To unsubscribe from the CCPEM list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCPEM&A=1
|