| layout | permalink |
|---|---|
default |
/guide/ |
Reproducibility in research is something that publishers, funding agencies, universities, research leaders and the general public worries about and much is being written about it. It is also something that researchers care deeply about - this lesson is typically one of the most popular lessons in the pre-workshop survey.
Even though most PhD students, postdocs and researchers (i.e. typical workshop participants) know about the importance of reproducibility in research, they often lack both a general overview of what different aspects there are to reproducibility, and the knowledge of specific tools that can be used for improving reproducibility.
Many participants may not adhere to good practices when organizing their projects, and the "Organizing your projects" episode is meant to encourage participants to structure their projects better. This may be obvious to some participants but it doesn't harm to preach to the choir.
Many participants do not share data or code, and in particular they are often unaware that research output other than published articles can be assigned digital object identifiers (DOIs). The "Sharing code and data" is meant to bring awareness about FAIRness and Open Science and to show how DOIs can be obtained for Git repositories..
Even though many participants know that code can have many dependencies (e.g. they may have experienced difficulties in getting other people's code to run), they often don't know or use good practices when it comes to recording dependencies. Most participants also don't use isolated environments for different projects and don't know why that can be important. The episode "Recording dependencies" tries to convey the importance of recording dependencies accurately for your projects, and shows how tools like conda can be used both as a package and software environment manager.
Many participants have heard about containers and find them interesting, but lack an understanding of how they work or how they can be used. The episode "Recording environments" introduces the concept of containers, and the optional episode "Creating and sharing a container image" goes into details.
Many participants use complicated series of computational steps in their research without realizing that this work falls into the category of "scientific workflows", and that there actually exist tools that help make such workflows reproducible. The episode "Recording computational steps" introduces the concept of scientific workflows, discusses various ways of managing workflows with varying degrees of reproducibility, and shows how tools like make and snakemake can be used to both simplify workflows and make them more reproducible.
By the end of this lesson, learners should:
- understand why it's important that research is reproducible
- define multiple levels of reproducibility and the multiple factors which can lead to irreproducible research
- be able to apply well organized directory structure for their project
- remember the FAIR principles
- be able to create a DOI for research output like code or data
- understand that code can have dependencies, and know how to create a requirements file
- describe how to use conda (or another tool) to manage isolated software environments
- have an idea about how containers work
- remember the pros and cons of containers for reproducibility
- understand that a series of computational steps can be defined as a workflow and that many tools exist to manage workflows
- describe the pros and cons of different methods of performing scientific workflows
- understand the benefits of using a tool like make or snakemake compared to bespoke scripts
- be able to create a simple Makefile or Snakefile to manage a workflow
Everyone knows that scientific results need to be reproducible, but not everyone is using appropriate tools to ensure this. Here we're going to get to know tools which help with preserving the provenance of data and reproducibility on different levels, ranging from workflow automation to software environment (containers).
Try to explain better what the different tools are useful for, but don't go into details. In this lesson we are not trying to gain expertise in the various tools and master the details but rather we want to give an overview and show that many tools exist and try to give participant the right feel for which set of tools to approach for which type of problem.
- Have you ever been unable to reproduce a scientific results from previous research by yourself or others?
- What technical solutions do you use to manage repetitive tasks and/or complicated workflows?
Depending on workshop schedule, this lesson can become short on time.
- The Snakemake episode is long, but the first steps in constructing the Snakefile can be covered quickly. Enough time should be devoted to the exercise though. Wildcards and pattern rules should be discussed, but the optional further topics can be skipped or just very briefly described
- Building the container from scratch (fetching the ubuntu image, installing all packages in it) can take several minutes. The instructor should in most cases do this before the workshop and only demonstrate how to launch a container based on the existing image
- The motivation can be gone through quickly. Researchers know the importance of reproducibility and we don't need to spend many words on it
- the body of a rule and the body of an input keyword need to be indented, but the number of spaces doesn't matter This works:
rule alldata:
input:
'processed_data/isles.dat',
'processed_data/abyss.dat'but this doesn't work:
rule alldata:
input:
'processed_data/isles.dat',
'processed_data/abyss.dat'not this:
rule alldata:
input:
'processed_data/isles.dat',
'processed_data/abyss.dat'Try to use good analogies to explain the concepts of containers, images and Dockerfiles.