Hi Dieter, You can use relion_stack_create. Do change _rlnReferenceImage to _rlnImageName in the star file of class average. Best wishes Kai On Wed, 2014-09-03 at 18:47 +0200, Dieter Blaas wrote: > Hi Kai, > thank you very much for this ample explanation! Just one final question: > How do I write out the images of the class averages? I can display them, > I can write out a star file referring to original image stacks but I > could not find a way of writing out an mrcs stack containing just the > image of each class average (as shown in 'Display'). Relion_preprocess > --operate_on star-file_of_class_averages.star --o > stack_of_class_average_images.mrcs does not work! > Best, Dieter > > ------------------------------------------------------------------------ > Dieter Blaas, > Max F. Perutz Laboratories > Medical University of Vienna, > Inst. Med. Biochem., Vienna Biocenter (VBC), > Dr. Bohr Gasse 9/3, > A-1030 Vienna, Austria, > Tel: 0043 1 4277 61630, > Fax: 0043 1 4277 9616, > e-mail: [log in to unmask] > ------------------------------------------------------------------------ > > Am 03.09.2014 16:11, schrieb Kai Zhang: > > Hi Dieter, > > > > Here's my experience to get rid of bad particles in the following. > > Normally, I tend to deal with different types of bad particles in > > different ways. Ice contamination, aggregation, carbon edge, obvious > > different protein and so on are regarded as 100% rubbish and should be > > excluded completely from data sets. To get rid of these rubbish, the > > best way is manually screening after auto-picking. You can low-pass and > > shrink all the particles, open the entire data set if possible and then > > quickly screen. It's tedious but works very well(I believe it's better > > than 2D classification); multiple rounds of 2D classification to get > > comparable results also requires quite a long time. Also you can use CCC > > or other values from auto-picking or projection match(if model > > available) to rank all particles and open them from low quality to high > > quality, then screening will be a bit faster. There are other bad > > particles simply because of the 'low-quality'(slightly distorted, drift, > > dried, over-exposure, falling apart, low contrast ...) rather than real > > rubbish. It's not easy to get rid of them by eye and 2D classification > > or 3D classification is more reliable than eye. However, ranking by CCC > > or phase error or maximum likelihood contribution doesn't really work > > well at beginning because of the low-quality model and possible bias. In > > 2D level you can do multiple rounds of classification. But I think it's > > faster to do it in a hierarchical way(classification and > > re-classification for each class). For the first round of entire data > > sets, I tend to use shrinked and low-passed particles. And later, use > > the original unfiltered particles. You might get rid of the very > > low-quality particles roughly by doing this. But it's better to keep > > suspicious classes that look like something in-between rubbish and good > > particles(they may contain valuable rare views). I normally don't > > completely get rid of these classes, but keep them for later 3D > > classification. In 3D level, I will also completely get rid of the > > really bad ones and keep suspicious classes like 2D. Once I get a > > better map, I will re-use these suspicious particles previously regarded > > as bad and re-rank the quality of all the particles. This can be done > > for multiple cycles. However if you have huge data sets, you don't need > > to do it in this way, just keep the very high-quality particles :-). > > > > I myself is always struggling at find ways to do it better, but I still > > hope this is helpful for you. > > > > Best wishes > > Kai > > > > On Tue, 2014-09-02 at 12:35 +0200, Dieter Blaas wrote: > >> Hi Kai, > >> thanks, 2D-classification is still running but so far I see that the > >> use of both parameters results in all classes being populated with > >> well-recognizable viral class averages. However, so far (iteration5) I > >> do not get any trash particles. Since I do not think that I got rid of > >> all (really all) 'bad' particles in the last run (#6), does this mean > >> that they are now incorporated into the classes and will reduce the > >> finally achieved resolution? I understand that there must be a tradeoff > >> between separating class averages and inclusion of less good particles. > >> However, what would be the best way to go for purification of the > >> dataset? Many selective runs (10 or more) to get rid of the trash and a > >> final run with these parameters set (for the sake of getting the class > >> averages)? > >> Best, Dieter > >> > >> ------------------------------------------------------------------------ > >> Dieter Blaas, > >> Max F. Perutz Laboratories > >> Medical University of Vienna, > >> Inst. Med. Biochem., Vienna Biocenter (VBC), > >> Dr. Bohr Gasse 9/3, > >> A-1030 Vienna, Austria, > >> Tel: 0043 1 4277 61630, > >> Fax: 0043 1 4277 9616, > >> e-mail: [log in to unmask] > >> ------------------------------------------------------------------------ > >> > >> Am 01.09.2014 00:20, schrieb Kai Zhang: > >>> Hi Dieter, > >>> > >>> Did you tried --only_flip_phases(or --ctf_intact_first_peak) and/or > >>> --strict_highres_exp? It helps a lot for uncleaned data sets with strong > >>> orientation preference. > >>> > >>> Kai > >>> > >>> On Sun, 2014-08-31 at 14:52 +0200, Dieter Blaas wrote: > >>>> Hi Sjors and all, > >>>> I am trying to get a reasonable number of 2D class averages by > >>>> running a 2D-classification on an old icosahedral virus dataset (about > >>>> 10,000 images, 50 classes). After each run I select the classes with > >>>> features recognizable as viral, remove the trash, and run it again. I am > >>>> now at run 5 and I am still getting only 2 classes recognizable as viral > >>>> projections (in sum about 99% of the images) and about 40 classes > >>>> containing between 1 and 20 'bad' images. I have tried T=1, 1.5, and 2 > >>>> and angular sampling 5 and 2.5 but regardless of these parameters it > >>>> seems to go on for ever (each new run finds new 'trash images' and I do > >>>> not get more than the 2 main classes (with the second one at about 3 % > >>>> only)). I admit that the dataset is skewed towards one single view but > >>>> shouldn't I get more than two main class averages ? Is there a way of > >>>> removing the trash and getting a reasonable number of views once for ever? > >>>> Thanks for hints, Dieter > >>>> > >>>> ------------------------------------------------------------------------ > >>>> Dieter Blaas, > >>>> Max F. Perutz Laboratories > >>>> Medical University of Vienna, > >>>> Inst. Med. Biochem., Vienna Biocenter (VBC), > >>>> Dr. Bohr Gasse 9/3, > >>>> A-1030 Vienna, Austria, > >>>> Tel: 0043 1 4277 61630, > >>>> Fax: 0043 1 4277 9616, > >>>> e-mail: [log in to unmask] > >>>> ------------------------------------------------------------------------ > > > > > > > >