Distribution and Reproducibility

It is recommended to store each workflow in a dedicated git repository of the following structure:

├── .gitignore
├── README.md
├── LICENSE.md
├── workflow
│   ├── scripts
|   │   ├── script1.py
|   │   └── script2.R
│   ├── rules
|   │   ├── module1.smk
|   │   └── module2.smk
│   ├── report
|   │   ├── plot1.smk
|   │   └── plot2.smk
│   └── envs
|   │   ├── tool1.smk
|   │   └── tool2.smk
├── config
│   ├── config.yaml
│   └── some-sheet.tsv
└── Snakefile

Then, a workflow can be deployed to a new system via the following steps

# clone workflow into working directory
git clone https://github.com/user/myworkflow.git path/to/workdir
cd path/to/workdir

# edit config and workflow as needed
vim config/config.yaml

# execute workflow, deploy software dependencies via conda
snakemake -n --use-conda

Importantly, git branching and pull requests can be used to modify and possibly re-integrate workflows. A cookiecutter template for creating this structure can be found here. Given that cookiecutter is installed, you can use it via:

cookiecutter gh:snakemake-workflows/cookiecutter-snakemake-workflow

Visit the Snakemake Workflows Project for best-practice workflows.

Integrated Package Management

With Snakemake 3.9.0 it is possible to define isolated software environments per rule. Upon execution of a workflow, the Conda package manager is used to obtain and deploy the defined software packages in the specified versions. Packages will be installed into your working directory, without requiring any admin/root priviledges. Given that conda is available on your system (see Miniconda), to use the Conda integration, add the --use-conda flag to your workflow execution command, e.g. snakemake --cores 8 --use-conda. When --use-conda is activated, Snakemake will automatically create software environments for any used wrapper (see Wrappers). Further, you can manually define environments via the conda directive, e.g.:

rule NAME:

with the following environment definition:

 - r
 - r=3.3.1
 - r-ggplot2=2.1.0

The path to the environment definition is interpreted as relative to the Snakefile that contains the rule (unless it is an absolute path, which is discouraged).

Snakemake will store the environment persistently in .snakemake/conda/$hash with $hash being the MD5 hash of the environment definition file content. This way, updates to the environment definition are automatically detected. Note that you need to clean up environments manually for now. However, in many cases they are lightweight and consist of symlinks to your central conda installation.

Conda deployment also works well for offline or air-gapped environments. Running snakemake -n --use-conda --create-envs-only will only install the required conda environments without running the full workflow. Subsequent runs with --use-conda will make use of the local environments without requiring internet access.

Running jobs in containers

As an alternative to using Conda (see above), it is possible to define, for each rule, a docker or singularity container to use, e.g.,

rule NAME:

When executing Snakemake with

snakemake --use-singularity

it will execute the job within a singularity container that is spawned from the given image. Allowed image urls entail everything supported by singularity (e.g., shub:// and docker://).

When --use-singularity is combined with --kubernetes (see Executing a Snakemake workflow via kubernetes), cloud jobs will be automatically configured to run in priviledged mode, because this is a current requirement of the singularity executable. Importantly, those privileges won’t be shared by the actual code that is executed in the singularity container though.

Combining Conda package management with containers

While Integrated Package Management provides control over the used software in exactly the desired versions, it does not control the underlying operating system. Here, it becomes handy that Snakemake >=4.8.0 allows to combine Conda-based package management with Running jobs in containers. For example, you can write

singularity: "docker://continuumio/miniconda3:4.4.10"

rule NAME:

in other words, a global definition of a container image can be combined with a per-rule conda directive. Then, upon invocation with

snakemake --use-conda --use-singularity

Snakemake will first pull the defined container image, and then create the requested conda environment from within the container. The conda environments will still be stored in your working environment, such that they don’t have to be recreated unless they have changed. The hash under which the environments are stored includes the used container image url, such that changes to the container image also lead to new environments to be created. When a job is executed, Snakemake will first enter the container and then activate the conda environment.

By this, both packages and OS can be easily controlled without the overhead of creating and distributing specialized container images. Of course, it is also possible (though less common) to define a container image per rule in this scenario.

The user can, upon execution, freely choose the desired level of reproducibility:

  • no package management (use whatever is on the system)
  • Conda based package management (use versions defined by the workflow developer)
  • Conda based package management in containerized OS (use versions and OS defined by the workflow developer)

Using environment modules

In high performace cluster systems (HPC), it can be preferable to use environment modules for deployment of optimized versions of certain standard tools. Snakemake allows to define environment modules per rule:

rule bwa:
        "bwa mem {input} | samtools view -Sbh - > {output}"

Here, when Snakemake is executed with snakemake –use-envmodules, it will load the defined modules in the given order, instead of using the also defined conda environment. Note that although not mandatory, one should always provide either a conda environment or a container (see above), along with environment module definitions. The reason is that environment modules are often highly platform specific, and cannot be assumed to be available somewhere else, thereby limiting reproducibility. By definition an equivalent conda environment or container as a fallback, people outside of the HPC system where the workflow has been designed can still execute it, e.g. by running snakemake –use-conda instead of snakemake –use-envmodules.

Sustainable and reproducible archiving

With Snakemake 3.10.0 it is possible to archive a workflow into a tarball (.tar, .tar.gz, .tar.bz2, .tar.xz), via

snakemake --archive my-workflow.tar.gz

If above layout is followed, this will archive any code and config files that is under git version control. Further, all input files will be included into the archive. Finally, the software packages of each defined conda environment are included. This results in a self-contained workflow archive that can be re-executed on a vanilla machine that only has Conda and Snakemake installed via

tar -xf my-workflow.tar.gz
snakemake -n

Note that the archive is platform specific. For example, if created on Linux, it will run on any Linux newer than the minimum version that has been supported by the used Conda packages at the time of archiving (e.g. CentOS 6).

A useful pattern when publishing data analyses is to create such an archive, upload it to Zenodo and thereby obtain a DOI. Then, the DOI can be cited in manuscripts, and readers are able to download and reproduce the data analysis at any time in the future.