Add a `specimen` model

[dyf]

Specimen is for keeping track of samples as they are created/subdivided by procedures.

Proposal:

• add a Specimen class to procedures.py
• and input_specimen and output_specimen to Procedure
• add a List[Specimen] to Procedures
• add specimen_id to EphysSession, Acquisition, and OphysSession.

[saskiad]

I think specimen_id only belongs in Acquisition. Ephys and Ophys Session still use intact subject.

[saskiad]

we need the procedure that creates specimens to be different ... not sure which procedure this is

[saskiad]

there should be very few procedures that have different input and output specimens...

[dyf]

@saskiad Sorry, missed these comments somehow.

In the past we separated the concept of the Subject from the brain. Right now we are treating those concepts as the same. Not sure how useful it is to separate them, but Specimen could do that.

More important is handling slicing and dissection appropriately, e.g.:

• Removing a section from an exaSPIM brain for spatial transcriptomics
• Slice physiology for patchseq sections

Here the input specimen would be the "brain" (we could just allow Subject for this), and the output would be a slice Specimen.

Other situations I have encountered:

• Perfusion and fixation are procedures that change the brain. We could treat these as separate Specimens. Maybe more headache than it's worth.
• Some experiments extract ROIs from sections for bulk sequencing. Specimens would be Brain -> Slice(s) -> ROIs.
• Patchseq is a variant of the above: Brain -> Slice(s) -> Cell (for reconstruction).

The Patchseq example is part of the origin story for the term cell_specimen in the past.

[dyf]

Summarizing offline conversation:

• Our data assets are for single recordings, therefore
• there will always be a single target specimen, therefore
• we do not have to capture a tree structure - just straight lineage back to the subject.
• This could be represented as
  1. a list of Specimens, each with a derivation_procedure
  2. a list of SpecimenProcedures, each with input and output specimens defined
  3. a nested dictionary - Specimen.derived_by_procedure.input_specimen.derived_by_procedure...
  4. a complete graph description of nodes (specimens) and edges (procedures).

I think I personally prefer (1).

[saskiad]

@saskiad Sorry, missed these comments somehow.

In the past we separated the concept of the Subject from the brain.

We still do. When we remove the brain from the animal, it is a specimen

[saskiad]

A specimen is a piece of tissue. If that tissue is sliced, it becomes new pieces of tissue with new specimen IDs. If the tissue is bathed in solution, it is the same piece of tissue.

[saskiad]

• Some experiments extract ROIs from sections for bulk sequencing. Specimens would be Brain -> Slice(s) -> ROIs.

Specimens are physical things. If we remove a chunk of tissue for sequencing, that's a new specimen. If we trace a cell on a computer image, that is not a specimen.

[dyf]

• Some experiments extract ROIs from sections for bulk sequencing. Specimens would be Brain -> Slice(s) -> ROIs.

Specimens are physical things. If we remove a chunk of tissue for sequencing, that's a new specimen. If we trace a cell on a computer image, that is not a specimen.

Agree. In my example the ROIs are dissected out.

[saskiad]

resolved by PR #170