Running Relion3.1
20200320 HMEDit
Add the following to your bashrc (you only need to add it once):
module use -a /well/strubi/relion/modulefiles
Save and close
Getting started
If you want to run jobs on GPUs e.g. for motion correction:
module load relion/3.1-gcc5.4.0-cuda8.0-libtiff4.0.10
->Motion Correction
/apps/strubi/motioncorr/2-1.1.0-gcc5.4.0-cuda8.0-sm61/MotionCor2
->GCTF, non-phase plate
/apps/strubi/gctf/Gctf-v1.06_sm_30_cu8.0_x86_64
->GCTF, phase plate
/apps/strubi/gctf/Gctf_v1.18_b2_sm37_cu8.0 (For phase plate data)
For CPU jobs, I would recommend:
module load relion/3.1-gcc5.4.0-skylake-libtiff4.0.10/bin/relion
Remember that if you have already module loaded a version of relion, you must unload this
module first before loading a new one.
e.g.
module unload relion/3.1-gcc5.4.0-skylake-libtiff4.0.10
module load relion/3.1-gcc5.4.0-cuda8.0-libtiff4.0.10
Bugs/Quirks
1. For refinement, I run into the following error when requesting an even
number of MPI procs:
==================
ERROR:
MlOptimiserMpi::initialiseWorkLoad: the number of MPI processes must be an odd number when gold-standard seperation is applied.
in: /well/strubi/relion/3.1-gcc5.4.0-skylake-libtiff4.0.10/src/ml_optimiser_mpi.cpp, line 575
ERROR:
MlOptimiserMpi::initialiseWorkLoad: the number of MPI processes must be an odd number when gold-standard seperation is applied.
=== Backtrace ===
This error can be quite buried in the error log file. Try grepping ERROR out of the file.
This is easily circumvented by entering an odd number for the number of MPI procs in the
relion GUI.
2. If you accidentally input the wrong file name for the 3D reference, the GUI has a tendency to crash immediately following attempted execution of the run.
In such cases where the gui crashes, in a similar way to ccp4i and phenix, the GUI will still be 'locked'. To open a fresh instance of a locked GUI, you must first delete the star file that tells relion that the gui is open, which is located in your relion instance home directory.
rm -r .relion_lock/
Observations
For virus structures, esp if capsids are relatively featureless, I often find it helpful
to pre-align particles (i.e. set number of classes to 1 in 3D classification tab) prior to
classification into multiple classes. In such cases, I often drop the search range when
moving on to classification.
CtfRefinement has had a substantial impact on my reported resolution values.
Following CTF refinement with beam tilt, per particle defocus, per micrograph astigmatism and per micrograph B-factor, I had an improvement in resolution (following a further round of refinement/post-processing) from 3.9 A to 3.3 A.
Useful links
RELION3.1 TUTORIAL: ftp://ftp.mrc-lmb.cam.ac.uk/pub/scheres/relion31_tutorial.pdf
PAPER: https://www.biorxiv.org/content/10.1101/823310v1
Sjors H.W. Scheres, Amyloid structure determination in RELION-3.1, 2019, doi: https://doi.org/10.1101/823310