Dear Alex,
This seems to be a very big box. You seem to run out of memory, so just
running on a bigger one should solve your problem. If you could run the
refinement on a cluster, the same nodes should be big enough, but make
sure you ask for enough RAM in your scheduler. Using or not using the MTF
doesn't affect the memory requirements.
HTH,
Sjors
> Dear Sjors
>
> when i try post-processing the model i got in 3D auto-refine step on an
> interactive node of a cluster, i.e. without
> submitting to the queueing system, i get memory allocation error:
>
> Executing: `which relion_postprocess` --i Refine3D/run1 --o
> Refine3D/postprocess  --angpix 1.12 --auto_mask  --inimask_threshold 0.01
> --extend_inimask 3 --width_mask_edge 3 --auto_bfac  --autob_lowres 10  &
> == Reading input half-reconstructions:
>   + half1-map:                     Refine3D/run1_half1_class001_unfil.mrc
>   + half2-map:                     Refine3D/run1_half2_class001_unfil.mrc
> == Perform auto-masking ...
>   + density threshold:             0.01
>   + extend ini mask:               3 pixels
>   + width soft edge:               3 pixels
>
> == Extending initial binary mask ...
> 2.52/2.52 min
> ............................................................~~(,_,">
> == Making a soft edge on the extended mask ...
> 2.98/2.98 min
> ............................................................~~(,_,">
> == Masking input maps ...
> +++ RELION: command line arguments (with defaults for optional ones
> between parantheses) +++
> ====== General options =====
>                                 --i : Input rootname, e.g. run1
>                            --angpix : Pixel size in Angstroms
>                   --o (postprocess) : Output rootname
> ====== Masking options =====
>                 --auto_mask (false) : Perform automated masking, based on
> a density threshold
>          --inimask_threshold (0.02) : Density at which to threshold the
> map for the initial seed mask
>               --extend_inimask (3.) : Number of pixels to extend the
> initial seed mask
>              --width_mask_edge (6.) : Width for the raised cosine soft
> mask edge (in pixels)
>                           --mask () : Filename of a user-provided mask
> (1=protein, 0=solvent, all values in range
> [0,1])
> ====== Sharpening options =====
>                            --mtf () : User-provided STAR-file with the
> MTF-curve of the detector
>                 --auto_bfac (false) : Perform automated B-factor
> determination (Rosenthal and Henderson, 2003)
>                --autob_lowres (10.) : Lowest resolution (in A) to include
> in fitting of the B-factor
>                --autob_highres (0.) : Highest resolution (in A) to include
> in fitting of the B-factor
>                   --adhoc_bfac (0.) : User-provided B-factor (in A^2) for
> map sharpening, e.g. -400
> ====== Filtering options =====
>        --skip_fsc_weighting (false) : Do not use FSC-weighting (Rosenthal
> and Henderson, 2003) in the sharpening process
>                     --low_pass (0.) : Resolution (in Angstroms) at which
> to low-pass filter the final map (by default at final
> resolution)
> ====== Expert options =====
>            --randomize_at_fsc (0.8) : Randomize phases from the resolution
> where FSC drops below this value
>             --filter_edge_width (2) : Width of the raised cosine on the
> low-pass filter edge (in resolution shells)
>                          --verb (1) : Verbosity
> Allocate: No space left
> File: ./src/multidim_array.h line: 927
>
> If i try to submit the job to queue (in *_submit.script I can allocate
> more RAM, which I assume is a problem here), i.e. in GUI:
> Post-processing->Running->Submit to Queue->yes,
> I get (even after pressing “Print command”):
> [3]+  Segmentation fault      relion
>
> I run RELION on a 64bit cluster with scientific linux:
> $ cat /proc/version
> Linux version 2.6.32-504.30.3.el6.x86_64 ([log in to unmask]
> <mailto:[log in to unmask]>) (gcc version 4.4.7 20120313 (Red Hat
> 4.4.7-11) (GCC) ) #1 SMP Tue Jul 14 11:18:03 CDT 2015
>
> Parameters that I use are set to default except for “Initial
> binarization threshold”, which I set to 0.01 based on checking contour
> level in Chimera.
>
> Are there any tricks to avoid it? Should more allocated RAM fix this
> problem?
>
> Thank you,
> Alex
>
> ps. Indeed I have just managed to run it after increasing allocated RAM,
> however, I get the following, suggesting that resolution became lower
> after refinement:
> $ tail -n 10 Refine3D/run1.out
>  Calculating gold-standard FSC ...
>  Maximization ...
> 5.61/5.61 hrs
> ............................................................~~(,_,">
>  Auto-refine: Refinement has converged, stopping now...
>  Auto-refine: + Final reconstruction from all particles is saved as:
> Refine3D/run1_class001.mrc
>  Auto-refine: + Final model parameters are stored in:
> Refine3D/run1_model.star
>  Auto-refine: + Final data parameters are stored in:
> Refine3D/run1_data.star
>  Auto-refine: + Final resolution (without masking) is: 12.1622
>  Auto-refine: + But you may want to run relion_postprocess to mask the
> unfil.mrc maps and calculate a higher resolution FSC
> DONE
> $ tail -n 10 Refine3D/run2pp.out
>   + correlation of fit:            0
> == Low-pass filtering final map ...
>   + filter frequency:              12.2927
> == Writing out put files ...
>   + Processed map:                 Refine3D/postprocess.mrc
>   + Processed masked map:          Refine3D/postprocess_masked.mrc
>   + Auto-mask:                     Refine3D/postprocess_automask.mrc
>   + Metadata file:                 Refine3D/postprocess.star
>   + FINAL RESOLUTION:              12.2927
> DONE
>
> Can it be because I am not using “MTF of the detector”? I am now
> playing with parameters for “extend binary map” and “add a soft
> edge” - is there anything else I should try out?
>
> Thank you
>
>


-- 
Sjors Scheres
MRC Laboratory of Molecular Biology
Francis Crick Avenue, Cambridge Biomedical Campus
Cambridge CB2 0QH, U.K.
tel: +44 (0)1223 267061
http://www2.mrc-lmb.cam.ac.uk/groups/scheres