How to handle phantoms and calibration data?

[dyf]

Characterizing instruments and rigs usual entails collecting some data without a traditional subject (a mouse, a human, etc). A couple examples:

• an empty gel to understand background noise levels
• beads to characterize a point spread function.

How should we represent this data? A couple options:

1. Allow data to have no Subject. This makes validation complicated, makes it easy to skip over the most important metadata. Leaving it empty also means there is no way to describe the phantom (e.g. "it's beads").
2. Make Subject more permissive (i.e. make sex and species) optional. This is not great as those fields are critical.
3. Inheritance. Make a base model (e.g. SubjectBase) that has enough fields to describe phantom data. Then make the current Subject inherit from it.
4. Composition. Isolate all of the non-generic fields appropriate for all subjects in an optional subfield (e.g. Subject.non_phantom_stuff).

I think (3) is probably best, but no strong feeling.

@saskiad @dbirman

Probably best handled in 2.0 regardless.

[saskiad]

There literally is no subject. I am very strongly against having fake subject information and I don't want to make subject mean something other than what it currently does.
Most likely there are also no procedures which is also a requirement for complete metadata.

I think our best option is to have a data level for calibration data and have our upload and completion checks be set to allow assets with this data level to pass without subject and procedures, but still require instrument(/rig) and acquisition. We should probably do that same for our simulated data.

Actually, we should probably think of it more like we think of the models. The metadata that we have for models is not the same as what we have for data. I'd put this in that category.

[dbirman]

DataLevel.CALIBRATION seems like a nice way to handle this, with validation logic modified to ignore missing subject/processing.json

[dbirman]

Although it's also occurring to me that there are real phantoms used in some situations for calibration (like MRI) where there really is a subject in some sense and that information should be tracked...