Dear Sjors and the rest of ccpem,
(1) Before I ran with my own data, I started testing on Relion 1.4 with the tutorial ribosome data downloaded from your website. To test the running time on our cluster, which has ten compute nodes, each with dual 12-core 2.6GHz opteron 6344 CPUs with 64GB of RAM. I tested the autorefine step with 4 nodes and it took 64 hrs to converge.
--------------------------------------------------------------------------------------------------------------------------------
ribosome: 835 particle with size of 200 px
computer: 4 nodes * 24 cores/(6 cores/job) = 16 parallel jobs, each with memory 15.6G
end: 25 itrs at resolution 20.7A
time: 64 hrs
--------------------------------------------------------------------------------------------------------------------------------
Then I started with my own data which was recorded on UltraScan CCD. This time I used 10 nodes and less RMB for each job. the autoefine converged after 24 hrs.
--------------------------------------------------------------------------------------------------------------------------------
UltraScan data: 1000 particles with size of 86 px
computer: 10 nodes * 24 cores/(1 cores/job) = 240 parallel jobs , each with memory 2.6G
end: 21 itrs at resoluton 32.0A
time: 24 hrs
--------------------------------------------------------------------------------------------------------------------------------
At last I ran another data which was recorded on Falcon II camera for the same sample. The tomograms are obviously noiser. This time the job didn't finished even after 4 days.
--------------------------------------------------------------------------------------------------------------------------------
Falcon data: 1400 particles with size of 74 px
computer: 10 nodes * 24cores/(2 cores/job) = 120 parallel jobs, each with memory 5.2G
end: unfinished yet after 8 itrs
times: 4 days
--------------------------------------------------------------------------------------------------------------------------------
So I was wondering if the autorefine step is data quality (contrast, noisy) dependent, and in my case what is the best Relion parameters to let the autorefine converge faster.
By the way I ran all three jobs with the recommended parameters in Relion interface: 7.5 degree for initial angle sampling, initial offset range 5, initial offset step 1 and local search from auto sampling 1.8 degree. No reference was applied for the refinement.
(2) My second question is if the user can provided initial angles for each subtomogram particles.
Thanks,
Bruce
University of Pittsburgh
|