Improving reconstruction with Frealign

Submitted by tarek on Wed, 2016-03-16 09:57

Forums

Frealign

Hi,

I started to use frealign just few weeks ago and play around with several parameters to improve the resolution.
By chance I realised that the FSC is somehow correlated to the number of cpus used for reconstruction.
The more cpus I use, i.e. the more partial stacks are created, the lower (means worse) the resolution gets.
Is that by any means intended? How many cpus do you usually use for your reconstructions?

Besides that, what other parameters do you consider for obtaining better resolutions?

Best,
tarek

niko Wed, 2016-03-16 18:06

This does not sound right. The results should be completely independent of the number of CPUs used. Not sure what you mean with "partial stacks".

tarek Thu, 2016-03-17 03:18

Hi Niko, thanks for your reply.

I have a stack of ~74000 particle images, those are separated more or less equally on the several nodes in parts e.g. sample_mult_reconstruct_r1_n1, sample_mult_reconstruct_r1_n14686, sample_mult_reconstruct_r1_n29371... sample_mult_reconstruct_r1_n58741.
To my understanding those parts are first reconstructed separately before merging all together for the final volume.
Please correct me if my assumption is wrong.
However, the correlation between resolution and jobnumber was reproducable in my case.

Best,
tarek

niko Thu, 2016-03-17 08:01

The different jobs do not calculate full reconstructions, they only calculate the sums that later are merged to calculate a full reconstruction. Have a look at the merge_3d.log file to see if all intermediate reconstruction files are read in and merged.