Label(s) for science and reproducible research

[nuest]

Just wanted to put here some thoughts for labels supporting the reproducible research efforts that many people use Docker for. Feedback welcome! If there is some support/consensus, I'd be happy to create a PR to discuss concrete details.

• repository-url could point to a repository or archive, where the image is preserved (long term), e.g. Zenodo, OSF, figshare, ..., e.g. org.label-schema.repository-url = "https://zenodo.org/record/45983"
• repository-doi as above, but with a DOI, e.g. org.label-schema.repository-doi = "10.5281/zenodo.45983"
• publication-doi is the DOI of a scientific publication, e.g. org.label-schema.publication-doi = "10.1145/2851581.2892483"

[lizrice]

This use in science is a bit outside my experience so sorry if these are dumb questions! Can you explain briefly how a DOI is different from a regular registry? Is the DOI ID a well-known format (so having it in a well-known label would allow tooling to do clever things with the DOI info)? Would this potentially be covered by the documentation label?

[nuest]

Yes, a DOI is a well-defined format for persistent identifers, see https://en.wikipedia.org/wiki/DOI and https://www.doi.org/factsheets/DOIKeyFacts.html

Tools could use it to retrieve exactly the right scientific article or public data repository an image belongs to. As Docker images as supplemental material are discussed, and scientific communities struggle with giving credit in a system that only values publications (= sth. with a DOI), it would be great if scientists could point users towards the paper that either describes the software in the container to make sure they get credit.

You could just put it in the documentation label and then try to extract it, with reasonable amount of work and some failures, I guess. A well-known label would be less prone to error.

publication-doi is probably the most important one on my suggestion list and could immediately be used by image authors.

Further alternatives (on second thought):

• One could also start with a single doi field, and continue to distinguish further once there is a need for it
• Have a field doi and support a list of DOIs in it, not distinguishing specific relations.

There is no consensus yet on how container images and items in a repository relate to each other, so that's more of a long term idea.

[nuest]

@lizrice apologies, I just re-read the text and the link to Wikipedia was missing, wrongly pasted. I updated my last post a little bit, let me know if I should describe anything in more detail!