Re: some curious things diagnostic plots in v0.83.1432

[Carsten Sachse]

Dear Mikel,

It looks like you ran into some small changes of the latest version 0.83.1432.

On Jan 21, 2015, at 10:28, Mikel Valle  wrote:

Dear Carsten,

 

1.-In the diag-plot containing the comparison between images and reprojections of the 3D map, now the averages of the images look nice. In previous version, in my hand, the averages were just mere single images (in most of the cases the same image with the highest cc). 

For high- and maximum resolution the angular samples is always increased by a factor of 5. This fact made the occupancy of the angular projection bins very small (much less than 20 or rather close to a few), i.e. the averages looked very noisy. For the purpose of clarity, in the revised version I summarized the angular bins as they are specified in the low- and medium-resolution refinement steps, e.g. every 4 degrees. As a result, these bins are more populated, the averages are computed from max. 20 segments and they look like a nice average. 

 

2.- In the FSC calculated with the masked structure there is an strange behavior with a local minimum at aprox 10 A. What could it mean?

The dip of the structural mask FSC is a result of the mask deconvolution. The procedure is described in Chen et al. 2013 (doi:10.1016/j.ultramic.2013.06.004). Briefly, the shown FSC is a composite FSC, i.e. the bit below 10 Å is without any mask by randomizing phases beyond the 10 Å and the bit below 10 Å is from the FSC deconvolved for the mask effect. The dip results from correlation spilling in Fourier shells of the phase-randomized FSC. The same effect can be found in other programs that apply this approach for example RELION. They simply recommend removing the poorly behaved data points for the reasons stated above. 

Best wishes,

Carsten

 

Best.

 

Mikel Valle

[Carsten Sachse]

Hi Mikel,

On Jan 22, 2015, at 12:11, Mikel Valle wrote:

Sorry to bother you again, but how does Spring handle the movie frames? 

Spring will use the combined frame average as a starting point for the frame processing refinement. In the movie processing, Spring will first generate a running average segment for alignment (specified by 'Frame average window size’) to improve signal-to-noise ratio for restrained x-y parameter determination. Second, the measured movements from a mean position will be averaged locally over a specified distance ('Frame local averaging distance') based on the assumption that adjacent helical segments experience similar movements in the same region of the micrograph.

does it work with selected helices? this way we may improve the alignment of frames based on entire pictures. 

 

The program ’segment’ allows to discard unwanted helices if you specify your desired subset of helices via the selection criteria and the associated spring.db. This way, ‘segment' will generate a stack with the selected helices only and the frame processing option in ’segmentrefine3d' becomes significantly less expensive to compute. 

If the newly computed alignments of the frames together with the applied helical restraints are more precise, you should see improvements in your 3D reconstruction.

Best wishes,

Carsten