RELION (for REgularised LIkelihood OptimisatioN) is a stand-alone computer program for Maximum A Posteriori refinement of (multiple) 3D reconstructions or 2D class averages in cryo-electron microscopy. It is developed in the research group of Sjors Scheres at the MRC Laboratory of Molecular Biology.
The underlying theory of MAP refinement is given in a scientific publication . If RELION is useful in your work, please cite this paper.
The more comprehensive documentation of RELION is stored on the Wiki
More extensive options and configurations are available here, but the outlines to clone and install relion for typical use are made easy through cmake.
On ubuntu machines, installing cmake, the compiler, and additional dependencies (mpi, fftw) is as easy as:
sudo apt install cmake build-essential mpi-default-bin mpi-default-dev libfftw3-dev
On other systems it is typically just as easy, you simply have to modify "apt" to the appropriate package manager. You will also need git, which is just as easy;
sudo apt install git
Once git and cmake are installed, relion can be easily installed through
git clone https://github.com/3dem/relion.git
cd relion
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=/where/to/install/ ..
make -j4
make install
(NOTES: "/where/to/install/.." is typically "/usr/local/relion". Make sure you create this directory beforehand. Installing to that location requires sudo, so in this case be sure to use "sudo make install" instead of "make install" in the last step.)
These steps will download the source-code, create a build-directory, then configure and build relion, and lastly install it to be generally available on the system.
If FLTK related errors are reported, please add -DFORCE_OWN_FLTK=ON to
cmake. For FFTW related errors, try -DFORCE_OWN_FFTW=ON.
RELION requires libtiff to read TIFF movies. Most Linux distributions have packages like
libtiff-dev or libtiff-devel. Note that you need a developer package. You need version 4.0.x
to read BigTIFF files. If you installed libtiff in a non-standard location, specify the location by
-DTIFF_INCLUDE_DIR=/path/to/include -DTIFF_LIBRARY=/path/to/libtiff.so.5.
RELION is intermittently updated, with both minor and major features. To update an existing installation, simply use the following commands
cd relion
git pull
cd build
make -j4
make install # (or "sudo make install")
If something went wrong, remove the build directory and try again from cmake.
Parts of the cryo-EM processing pipeline can be very computationally demanding, and in some cases special hardware can be used to make these faster. There are two such cases at the moment;
- Since RELION-2: Use one or more PGUs, or graphics cards. RELION only supports CUDA-capable GPUs of compute capabilty 3.5 or higher.
- Since RELION-3: Use the vectorized version. RELION only supports GCC and ICC 2018.3 or later.
There are more benefits than speed; the accelearated versions also have a decreased memory footprint. Details about how to enable either of these options is listed below.
If you are using intels compiler (icc), you cannot have both these features in the same binary at the moment. Not all tools are GPU-accelerated or vectorized.
Tools that are GPU-accelerated:
- relion_refine
- relion_autopick
To build with support for CUDA-accelerated kernels in addition to the original CPU version, build by setting CUDA=ON and CudaTexture=ON/OFF:
cd build
rm -r *
cmake -DCUDA=ON -DCudaTexture=ON ...
make -j4
make install
If you run relion_refine with a the "--gpu" flag, you will run the accelerated CUDA version of the kernels. If you leave out the "--gpu" flag, it will run the original CPU version.
Tools that are CPU-accelerated (vectorized):
- relion_refine
To build with support for CPU-accelerated kernels in addition to the original
CPU version, build by setting ALTCPU=ON
cd build
rm -r *
cmake -DALTCPU=ON ..
make -j4
make install
This will require the TBB (Threading Building Blocks) library. RELION will look for TBB and fetch+install it if it cannot find it on your system. You can force this behaviour (and make sure you are using the latest version) by adding
-DFORCE_OWN_TBB=ON
In addition, you can make use the Intel(R) Math Kernel Library (Intel(R) MKL). This is optional ( but will scale better with increased threads). Add this by:
-DMKLFFT=ON
If you run relion_refine with a the "--cpu" flag, you will run the accelerated version. If you leave it the original CPU version will be run. You should use this flag if you can, unless you want to verify old runs or behaviour.
With the Plasmodium ribosome benchmark noted on the RELION website, GCC 7.3 hardware-optimized builds appear to run about 3x slower than those built with Intel(R) Parallel Studio XE 2018 Cluster Edition (for reasons still under investigation).
To build with the Intel compiler:
source /opt/intel/impi/<version>/intel64/bin/mpivars.sh intel64cd build
source /opt/intel/compilers_and_libraries_<version>/linux/bin/compilervars.sh intel64
source /opt/intel/compilers_and_libraries_<version>/linux/mkl/bin/mklvars.sh intel64
rm -r *
CC=mpiicc CXX=mpiicpc cmake -DALTCPU=ON -DMKLFFT=ON ..
make -j4
make install
The Intel(R) Compiler (icc) is not free. However, if you want to run a version of RELION built with the Intel compiler, you can download the runtime libraries for free:
- The Intel C++ Compiler libraries (including TBB)
- The Intel Math Kernel Libraries and Intel MPI
- Download at least "Intel(R) Math Kernel Library (Intel MKL)" and "Intel MPI Library (Linux Package)" (make sure the product "Intel(R) Performance Libraries for Linux*" is selected at the top)
The above needs to be installed in a location that is accessible to all machines running the icc-compiled binary.
Best performance seems to be seen when the pool size (--pool) is roughly the same as the number of threads (--j). Lowering the pool size may also decrease the memory used by a process or rank. When running multi-node, 4 MPI ranks per node seems to also work well (so --j and --pool should be set to the total number of threads in the machine divided by 4), although this may also depend on the data set. Systems with 256GB or 512GB are recommended for the CPU-accelerated kernels.