workflowr VS data-science-development-project-template

I would start by saying "bioinformatics world" is a very broad term. Most of it does not involve managing multiple models, which is what MLflow seems to be for. Most work is cleaning the data and interpreting the results, which is highly project-specific. Something like workflowr is generally more appropriate, but even that is an overkill for most people.

Targets stores all your data serialised on disk, and only loads them in as needed by the dependencies. For a standardised folder structure, you can take inspiration from workflowr.

Bioinf has a lot of biologists who have transitioned into more technical/coding focused roles, so you'll find there's not a lot of engineering workflow standards out there compared to DS or SWE. As others have said, snakemake is the most common, but thats just a pipeline managment tool, it doesn't manage data or outputs. I personally use DVC for data and pipeline management (and include jupyter and papermill to make it all work), although I haven't yet gotten onboard with their experiments feature (which is what would manage different parameters and figures/results beyond versioning). I looked into MLflow and some other options when I was getting started (I do tool development and bioinf analysis), but I wanted data versioning to ensure experiment reproducibility (kind of a critcal part of science IMO), and many of the other solutions like Airflow (common in DS industry) seemed to be overkill for smaller bioinfo projects. DVC meets the requirements and I like it in concept, although in practice there have been many updates that have been a bit of a pain to keep up with/integrate. I've got a bioinfo/ds project template on github that roles together git, conda, DVC, jupyter and papermill to ensure experiment reproducibility, and is setup as a template that can be deployed with cookiecutter - check it out if you like.