GeRelion is a GPU-enhanced parallel implementation of single particle cryo-EM image processing. Developed by Huayou Su, Wen Wen, Xiaoli Du, Xicheng Lu,Dongsheng Li from National University of Defense Technology and Maofu Liao from Harvard Medical School. For the current version, we implemented the most time-comsuming components of RELION with GPU, including 3D classificaion and auto-refine.

The implementation is built upon RELION (REgularised LIkelihood Optimisation). RELION is a stand-alone computer program that employs an empirical Bayesian approach to refinement of (multiple) 3D reconstructions or 2D class averages in electron cryo-microscopy (cryo-EM). It is developed in the group of Sjors Scheres at the MRC Laboratory of Molecular Biology. More detailed information about RELION can be found

The original RELION codes for data read/write and MPI communication are unmodified, and the flow control of progressive processing in the original RELION is kept. The *.cu source files and their corresponding header files in directory src are developed by GeRelion team. The *.cpp source files and their corresponding header files in directory src are from relion-1.3. Some of the *.ccp and *.hpp files from relion are modified by GeRelion team.

1. Hardware requirements To use the program GeRelion, you should have the CUDA-enabled GPUs from NVIDIA Corporation. GeRelion prefers NVIDIA GPUs with arch_sm equals 3.5 or prior.
2. Software requirements The CUDA Toolkit must be installed successfully. In addition, the thrust library and MPI pragraming interface should be installed.
3. The thiry party library fftw and fltk is required, both of them are put int the folder external.

1. Change the directory of your project in Makefile, check line 1. CUDA_DIR := path to your cuda directory installed (such as /usr/local/cuda-7.5)

2. Move to your GeRelion project folder and run the script INSTALL.sh. The compilation will last for several minites due to the usage of thrust library. The executable binary file will be put in bin directory of the project.

Replace “$GERELION_HOME” with the actual path of your GeRelion directory. If csh or tcsh, add the following lines to your ~/.cshrc.

export LD_LIBRARY_PATH=$GERELION_HOME/lib:$LD_LIBRARY_PATH export PATH=$GERELION_HOME/bin:$PATH

If bash, add the following lines to your ~/.bashrc:

setenv LD_LIBRARY_PATH $GERELION_HOME/lib:$LD_LIBRARY_PATH setenv PATH $GERELION_HOME/bin:$PATH

The current version of GeRelion only supports job submission via command line (GUI will be implemented soon). GeRelion uses the same parameters as RELION, so you can use RELION GUI "print command" to get all the necessary parameters. You can also skip GUI, and use samrelion.py to conveniently print out all parameters. Note that, without using RELION pipeline, samrelion.py can generate initial star files, analyze and modify data star file, generate the star files with particular class numbers, merge small groups, etc.

One GeRelion-specific parameter (--mode) is used to indicate running mode of GeRelion: 0 for CPU, and 1 for GPU.

A step-by-step instruction to test GeRelion using the TRPV1 data set:

1. Download the TRPV1 particle stack tv1_f01-30.mrc, and rename it as “tv1.mrcs”.
2. Download the TRPV1 3D reconstruction emd_5778.map, and rename it as “tv1.mrc”
3. Download this star file tv1.star.
4. The following commands assume that you have 2 GPU nodes (“node01” and “node02”), each containing 4 GPU cards. “$GERELION_HOME” should be replaced by the actual path of your GeRelion directory.

Run the following command for 3D auto-refine (with C4 symmetry):
mpirun --np 9 --host node01,node02 $GERELION_HOME/bin/gereline_refine_mpi --o REF01 --auto_refine --split_random_halves --i tv1.star --particle_diameter 200 --angpix 1.2156 --ref tv1.mrc --firstiter_cc --ini_high 60 --ctf --ctf_corrected_ref --flatten_solvent --zero_mask --oversampling 1 --healpix_order 2 --auto_local_healpix_order 4 --offset_range 5 --offset_step 2 --sym C4 --low_resol_join_halves 40 --norm --scale --j 1 --memory_per_thread 4 --dont_combine_weights_via_disc --mode 1

Run the following command for 3D classification (3 classes, without symmetry):
mpirun --np 9 --host node01,node02 $GERELION_HOME/bin/gereline_refine_mpi --o CLS01 --i tv1.star --particle_diameter 200 --angpix 1.2156 --ref tv1.mrc --firstiter_cc --ini_high 60 --ctf --ctf_corrected_ref --iter 25 --tau2_fudge 4 --K 3 --flatten_solvent --zero_mask --oversampling 1 --healpix_order 2 --offset_range 5 --offset_step 2 --sym C1 --norm --scale --j 1 --memory_per_thread 4 --dont_combine_weights_via_disc --mode 1

If you only have one node with 4 GPU cards, the commands are:
mpirun --np 5 $GERELION_HOME/bin/gereline_refine_mpi ... ...

GeRelion is released under the terms of the GNU General Public License as published by the Free Software Foundation.

Contact: Huayou Su [email protected]