Re: segementrefine3d

[Carsten Sachse]

Dear Mikel,

On 13 Mar 2014, at 15:46, Mikel Valle <mva...@cicbiogune.es> wrote:

Dear Dr. Sachse,

 

I am writing from a research center in the north of Spain. We are currently using Spring for the image processing of a helical virus. The package is amazing. We are newcomers to the helical symmetry and I have some doubts about the refinement. Could you please help us out? There are few questions:

 

1.-When preparing the segments for the refinement we use the "convolve" option on the segment command. The output seems to have lost most of the contrast at low frequencies, and the helices are barely visible. I´ve never worked before with convolution at 2D (just CTF correction on 3D data, or phase flipping in images), so, I do not know whether the images should look so poorly contrasted.

You are aware that we tested this approach multiple times (http://www.sachse.embl.de/emspring/publications.html) - the approach and our implementation works correctly. The image appearance is dominated by the low-resolution contrast that is actually being down-weighted relative to higher resolution contrast when you convolve the images. But often for helices you have sufficient high-resolution signal to base your alignment on. If you think this limits your analysis of your specimen you can choose to phase-flip the images instead. 

 

2.-In the segmentrefine3d parameter file it asks for the "Image alignment size" and the "Step size of segmentation". I thought that those two are set in the segmentation itself. Since the step size is crucial, I would like to get it clear.

The step size parameter should remain fixed as set in the segmentation. The maximum image alignment size is determined by your segmentation procedure. In the refinement you can choose to make this smaller. This way you may improve compensation of filament curvature but you may also worsen your alignment accuracy because decrease the amount of your signal in the image used for alignment.   

 

3.-About the "Assemble refinement strategy". It seems that one needs to select all the aims up to desired level, ie, if I select HR, the programs does not progress until I also pick LR, MR. Is the program organized this way?

Yes, you should select LR, MR in order to go HR. This gives you maximum flexibility of arranging your refinement strategy. You are aware that HR for example requires some starting parameters because it does not search the entire range of Euler angles anymore etc. You can assemble your own refinement strategy however you want. We found that the defaults are a good compromise at reasonable computational cost that still give you best results.

 

4.-the "3D CTF correction" option. This is puzzling to me. Since the segments are CTF corrected , what does this step add?. Also, the "3D CTF correction intensity" is a mystery to me. For the case of maximum resolution which would be the correct option?

Please, refer to our JMB paper (http://dx.doi.org/10.1016/j.jmb.2007.05.088) for the theory behind it. This is essentially an amplitude correction after convolution of the images by the CTF. For the intensity, we found to stick to 'low' as perfectly reasonable. In my experience, there is not much adjustment needed.

 

5.-In using movie frames from direct detection cameras. It seems that we run first the segementrefine3d with the summed micrographs, then go back to the segment command, and extract the images from the movie-framed files. Does the program align each segment/helix? Is there any way to evaluate its performance? In the commentary of the option it is said: This option is intended for movie processing of frames from direct detectors. Prior to this option run segmentrefine3d using the combined average of all frames. With the outputof this option you can re-run segmentrefine3d with movie refinement. The output of this option is:

The output is ....  I guess that a new file with aligned segments, and then we go to refinement again?

Yes, you need to provide a completed refinement.db where single micrographs, i.e. the sum of the frames, were used. Based on this previous refinement, you use 'segment' again to have all individual frames present in your frame data set. With the frame data set, you only perform a local refinement, e.g. for maximum resolution to improve the reconstruction. In Spring there are many statistics apart from the FSC to judge the performance of your reconstruction in general, e.g. in-plane rotation matches (polarity if applicable), forward x-shift difference etc. We have tested this procedure with TMV images and it does improve the map. 

Best wishes,

Carsten

 

I am really sorry to bother you, but I need some clarifications to proceed properly.

 

Thanks very much for your time.

 

Best.

 

Mikel Valle, PhD

Structural Biology Unit, CICbioGUNE

Biscay Technology Park, Bld 800

48610, Derio, Spain

 

 

 

 

 

 

This e-mail comes from CIC bioGUNE. The e-mail and any files transmitted with it are confidential and intended solely for the use of the individual or entity to which they are addressed. Any unauthorised dissemination or copying of this e-mail or its attachments, and any use or disclosure of any information contained in them, is strictly prohibited and may be illegal. If you have received this e-mail in error, please notify or telephone +34 944 06 13 00 us and delete it from your system. 

Please consider the environment before printing this email