Open Science

The opposite of “open” isn’t “closed”. The opposite of “open” is “broken”.

— John Wilbanks

Free sharing of information might be the ideal in science, but the reality is often more complicated. Normal practice today looks something like this:

• A scientist collects some data and stores it on a machine that is occasionally backed up by her department.

• She then writes or modifies a few small programs (which also reside on her machine) to analyze that data.

• Once she has some results, she writes them up and submits her paper. She might include her data – a growing number of journals require this – but she probably doesn’t include her code.

Time passes.

• The journal sends her reviews written anonymously by a handful of other people in her field. She revises her paper to satisfy them, during which time she might also modify the scripts she wrote earlier, and resubmits.

• More time passes.

• The paper is eventually published. It might include a link to an online copy of her data, but the paper itself will be behind a paywall: only people who have personal or institutional access will be able to read it.

For a growing number of scientists, though, the process looks like this:

• The data that the scientist collects is stored in an open access repository like figshare or Zenodo, possibly as soon as it’s collected, and given its own Digital Object Identifier (DOI). Or the data was already published and is stored in Dryad.

• The scientist creates a new repository on GitHub to hold her work.

• As she does her analysis, she pushes changes to her scripts (and possibly some output files) to that repository. She also uses the repository for her paper; that repository is then the hub for collaboration with her colleagues.

• When she’s happy with the state of her paper, she posts a version to arXiv or some other preprint server to invite feedback from peers.

• Based on that feedback, she may post several revisions before finally submitting her paper to a journal.

• The published paper includes links to her preprint and to her code and data repositories, which makes it much easier for other scientists to use her work as starting point for their own research.

This open model accelerates discovery: the more open work is, the more widely it is cited and re-used. However, people who want to work this way need to make some decisions about what exactly “open” means and how to do it. You can find more on the different aspects of Open Science in this book.

This is one of the (many) reasons we teach version control. When used diligently, it answers the “how” question by acting as a shareable electronic lab notebook for computational work:

• The conceptual stages of your work are documented, including who did what and when. Every step is stamped with an identifier (the commit ID) that is for most intents and purposes unique.

• You can tie documentation of rationale, ideas, and other intellectual work directly to the changes that spring from them.

• You can refer to what you used in your research to obtain your computational results in a way that is unique and recoverable.

• With a version control system such as Git, the entire history of the repository is easy to archive for perpetuity.

• Making Code Citable

Anything that is hosted in a version control repository (data, code, papers, etc.) can be turned into a citable object. You’ll learn how to do this in the later episode on Citation.

• How Reproducible Is My Work?

Ask one of your labmates to reproduce a result you recently obtained using only what they can find in your papers or on the web. Try to do the same for one of their results, then try to do it for a result from a lab you work with.

• How to Find an Appropriate Data Repository?

Surf the internet for a couple of minutes and check out the data repositories mentioned above: Figshare, Zenodo, Dryad. Depending on your field of research, you might find community-recognized repositories that are well-known in your field. You might also find useful these data repositories recommended by Nature. Discuss with your neighbor which data repository you might want to approach for your current project and explain why.

• How to Track Large Data or Image Files using Git?

Large data or image files such as .md5 or .psd file types can be tracked within a github repository using the Git Large File Storage open source extension tool. This tool automatically uploads large file contents to a remote server and replaces the file with a text pointer within the github repository.

Try downloading and installing the Git Large File Storage extension tool, then add tracking of a large file to your github repository. Ask a colleague to clone your repository and describe what they see when they access that large file.