Curate annotation transfer from MTG to Siletti taxonomies

[dosumis]

Depends on

• [x] #3
• [x] #17

[dosumis]

MTG taxonomy is here: https://github.com/hkir-dev/human-neocortex-middle-temporal-gyrus Annotation transfer mappings (on name) are here: https://docs.google.com/spreadsheets/d/1cC6L9oOGk7tb8pFtBOtol-caYculMdkALcPF-MxhRmM/edit#gid=1294367862 (Column F) Relevant part of CAS:

transferred_annotations (list)

• transferred_cell_label (string): Transferred cell label.
• source_taxonomy (string): PURL of source taxonomy.
• source_node_accession (string): accession of node that label was transferred from.
• algorithm_name (string): .
• comment (string): Free text comment on annotation transfer.

Example:

label: MGE_241
labelset: Cluster
accession: CS202210140_242
transferred_annotations: 
    - transferred_cell_label: sst
       source_taxonomy:  # we need to publish to PURL
       source_node_accession: 8fa477a378. # Publishing will allow us to generate a namespaced accession
       algorithm_name:  # N/A - they used a bespoke method
       comment: "We performed PCA (50 components) on our full dataset, trained a random forest classifier (scikit-learn, class_ weight=‘balanced’, max_depth=50) on the MTG labels, and then predicted labels for all cells. We labeled each cluster with the mode of its constituent cells if two conditions were met: more than 0.8 of predicted labels matched the mode, and the mean probability of these pre- dictions was greater than 0.8."

Strategy:

1. Work branches on Siletti repos # Turns out that this breaks TDT (presumably unless everything worked on is in .gitignore)
2. Install CAS-tools. Map AT on names (tracking anything that doesn't map).
3. Generate CAS objects & insert into
4. Run validator built into CAS.