FAIRness of research workflows

As mentioned in the introduction, an overarching goal is to make research software FAIR. In this section, we therefore want to discuss possible aspects of FAIRness associated with the code and data that are used in or produced by scientific inquiry. A scientific workflow may consist of a variety of different pieces of software as well as input data, which makes it nontrivial to publish it in a comprehensive, reusable and reproducible way.

Increasing transparency

By publishing the data and software used in scientific investigations, researchers give their peers the opportunity to retrace the steps that were necessary to arrive at the conclusions drawn in a publication. This significantly increases the transparency of their research and besides this, it allows others to reuse the data and software in proceeding investigations.

Reusable data

Integration of externally published research data in a research project may be cumbersome if the data is not well-documented or self-explanatory. To mitigate this, data should be published in standard formats that are widely used in the community to guarantee their seamless integration into other projects. Moreover, rich metadata should be provided, which explains the contents of the data in detail and in a machine-readable format.

Reusable software

Source code used to process or produce data should be published such that other researchers can easily reconstruct how the data was generated or analyzed. Besides a license which states terms and conditions of use, it is again important to publish metadata alongside the code, which state the requirements of the software on the compute environment, that is, compiler or interpreter versions, third-party libraries, hardware requirements etc.

Ideally, this code expresses only project-specific tasks, reusing established packages for generic functionality, whenever possible, instead of reimplementing it. This reduces the risk of bugs and makes it easier for other researchers to understand the code, who possibly are familiar with the syntax of the used packages. If generic, possibly reusable functionality has to be implemented, we encourage researchers to publish these parts separately in the form of a code repository and/or software package for others to use independently of the project-specific code. This, of course, requires competence in sustainable software development or the help of research software engineers.

Reproducible research workflows

A scientific workflow may involve many different processes that are executed in a particular order while exchanging data. With increasing number of processes, retracing the workflow logic can become rather cumbersome. Using workflow tools or languages to describe the logic and the flow of data provides the means to communicate this in a more comprehensible way. Moreover, some tools provide the possibility to visualize the workflow in a directed acyclic graph (DAG), depicting inputs, outputs and metadata of the processes and how they are interconnected. In addition to the graphical visualization of the provenance graph, the workflow tool may also automatically track and record the full data provenance in a database. Besides this, other researchers may easily extend or change the workflow by adding or exchanging processes.

In order for the workflow to be reproducible, metadata must be provided that state the requirements on the compute environment and all of its dependencies, such that users can prepare their system for the workflow execution. Ideally, workflow tools automatically track the involved dependencies and/or provide the means to encapsulate the environment (e.g. using container solutions) such that it can be distributed.