Installation
The latest versions of MDAnalysis can be installed using conda or pip. Currently, the conda releases only support serial calculations. If you plan to use the parallel OpenMP algorithms, you need to install MDAnalysis with pip and have a working OpenMP installation.
MDAnalysis has a separate test suite MDAnalysisTests that is required to run the test cases and examples. The test files change less frequently, take up around 90 MB of space, and are not needed for daily use of MDAnalysis. However, they are often used in examples, including many in this User Guide. If you are not interested in developing MDAnalysis or using the example files, you most likely don’t need the tests. If you want to run examples in the User Guide, install the tests. The tests are distributed separately from the main package.
Note
If you are installing on Windows, you must have Microsoft Visual C++ 14.0 installed. If your installation fails with the error message:
error: Microsoft Visual C++ 14.0 is required. Get it with “Build Tools for Visual Studio”: https://visualstudio.microsoft.com/downloads/
Try installing Build Tools for Visual Studio from https://visualstudio.microsoft.com/downloads/ (scroll down to the Tools for Visual Studio section).
If you encounter any issues following these instructions, please ask for help on GitHub Discussions (Installation).
conda
If you use conda to manage your Python environment, we highly recommend creating a new environment for MDAnalysis.
This will ensure that you have a clean installation of MDAnalysis and its dependencies, and will not interfere with other packages you may have installed.
We further recommend that you install and use mamba, a faster drop-in replacement for conda.
conda create --name mdanalysis
conda activate mdanalysis
conda install -c conda-forge mamba
To install the latest stable version of MDAnalysis via conda, use the following command. This installs all dependencies needed for full analysis functionality (excluding external programs such as HOLE):
mamba install -c conda-forge mdanalysis
To upgrade:
mamba update mdanalysis
To install the tests:
mamba install -c conda-forge MDAnalysisTests
If you intend to use MDAnalysis in JupyterLab, you will have to install an extra package for the progress bar in analysis classes:
conda install -c conda-forge nodejs
jupyter labextension install @jupyter-widgets/jupyterlab-manager
pip
The following command will install or upgrade the latest stable version of MDAnalysis via pip, with core dependencies. This means that some packages required by specific analysis modules will not be installed.
pip install --upgrade MDAnalysis
If you need to install a fully-featured MDAnalysis, add the analysis tag. As with conda, this will not install external programs such as HOLE.
pip install --upgrade MDAnalysis[analysis]
To install/upgrade tests:
pip install --upgrade MDAnalysisTests
If you intend to use MDAnalysis in JupyterLab, you will have to install an extra package for the progress bar in analysis classes:
pip install nodejs
jupyter labextension install @jupyter-widgets/jupyterlab-manager
Development versions
To install development versions of MDAnalysis, you can compile it from source. In order to install from source, you will need numpy and cython. See Creating a development environment for instructions on how to create a full development environment.
git clone https://github.com/MDAnalysis/mdanalysis
cd mdanalysis
# assuming you have already installed required dependencies
pip install -e package/
And to install the test suite:
pip install -e testsuite/
Testing
The tests rely on the pytest and numpy packages, which must also be installed. Run tests with:
pytest --disable-pytest-warnings --pyargs MDAnalysisTests
All tests should pass (i.e. no FAIL, ERROR); SKIPPED or XFAIL are ok. If anything fails or gives an error, ask on GitHub Discussions or raise an issue.
Testing MDAnalysis can take a while, as there are quite a few tests. The plugin pytest-xdist can be used to run tests in parallel.
pip install pytest-xdist
pytest --disable-pytest-warnings --pyargs MDAnalysisTests --numprocesses 4
Custom compiler flags and optimised installations
You can pass any additional compiler flags for the C/C++ compiler using the extra_cflags variable in setup.cfg.
This allows you to add any additional compiler options required for your architecture.
For example, extra_cflags can be used to tune your MDAnalysis installation for your current architecture using the -march, -mtune, -mcpu and related compiler flags.
Which particular compiler flags to use depends on your CPU architecture. An example for an x86_64 machine would be to change the line in setup.cfg as follows:
- #extra_cflags =
+ extra_cflags = -march=native -mtune=native
Use of these flags can give a significant performance boost where the compiler can effectively autovectorise.
Be sure to use the recommended flags for your target architecture. For example, ARM platforms recommend using -mcpu instead of -mcpu, while
PowerPC platforms prefer both -mcpu and -mtune.
Full dicussion of the these flags is available elsewhere (such as here in this wiki or in this ARM blog post) and a list of supported options should be provided by your compiler. The list for GCC is provided here.
Warning
Use of these compiler options is considered advanced and may reduce the binary compatibility of MDAnalysis significantly, especially if using -march, making it usable only on a matching CPU architecture to the one it is compiled on. We strongly recommend that you run the test suite on your intended platform before proceeding with analysis.
In cases where you might encounter multiple CPU architectures (e.g. on a supercomputer where the login node and compute node have different architectures), you should avoid changing these options unless you are experienced with compiling software in these situations.
Additional datasets
MDAnalysisData is an additional package with datasets that can be used in example tutorials. You can install it with conda or pip:
# conda
conda install -c conda-forge mdanalysisdata
# pip
pip install --upgrade MDAnalysisData
This installation does not download all the datasets; instead, the datasets are cached when they are first downloaded using a Python command.