load_deepcellseek
datasets.load_deepcellseek(cell_type='Broad Cell type')
Load the bundled DeepCellSeek (Briefings 2025) annotation agreement scores.
The DeepCellSeek benchmark scores 11 LLM endpoints and the three classical annotators CellMarker2.0, SingleR and ScType on the same ontology-aware agreement metric as Hou and Ji (2024), so it reuses the two annotation cards and the :class:GPTCelltype container. Reads deepcellseek2025_agreement.csv and reduces it to a method by dataset by metric tensor (mean agreement and full match rate). The classical method labels match load_gptcelltype so the two benchmarks can be compared on their shared methods.
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
cell_type |
str | Which type rows to keep, "Broad Cell type" by default to match the broad-annotation scope of load_gptcelltype. Pass "Subtype" for the subtype rows, or "all" to keep every row. |
'Broad Cell type' |
Returns
| Type | Description |
|---|---|
| GPTCelltype |