# Cell-type & tissue-niche annotation — human expert reference

8 human scientists annotated the **same** developing-human-heart MERFISH dataset
(228,633 cells × 238 genes). Identities are removed; experts are numbered **1–8** (this
numbering is independent of the panel-design task). Per-expert methodology is in
[`workflows.csv`](workflows.csv).

## Files

| File | Contents |
| --- | --- |
| `workflows.csv` | `id, cell_type_workflow, niche_workflow` — each expert's approach |
| `combined_annotations_anonymized.h5ad` | All 8 experts (anonymized) **+ model/baseline predictions** |
| `human_annotations_anonymized.h5ad` | Human experts only (model/baseline columns dropped) |
| `per_expert_raw/expert{N}.h5ad` / `.csv` | Each expert's original file, anonymized (native schema) |
| `per_expert_raw/expert7_niche.h5ad` | Expert 7's tissue-niche annotation (separate source file) |

## Combined object structure

Both combined `.h5ad` files share one cell index and embeddings:

- `X` — log1p-normalized expression (238 genes)
- `layers['raw_count']` — raw counts
- `obsm` — `X_pca`, `X_umap`, `spatial`

**Per-expert annotation columns** (N = 1..8):

```
cell_type_tier1_expert{N}        cell_type_tier2_expert{N}     [cell_type_tier3_expert{N}]
tissue_niche_tier1_expert{N}     tissue_niche_tier2_expert{N}
cell_type_expert{N}              tissue_niche_expert{N}        # consolidated single-label
```
Tier 3 is present only for experts who provided it (cell type: experts 2, 6, 7; niche: expert 7).
Expert 6 additionally has `cell_type_main_expert6`.

**Reference / shared columns:** `cell_type`, `tissue_niche` (consensus labels),
plus technical fields (`sample_id`, `batch`, `n_counts`, `leiden`, and cluster features).

**Model/baseline columns** (only in `combined_annotations_anonymized.h5ad`):
`cell_type_agent`, `tissue_niche_agent`, `cell_type_gpt`, `cell_type_sctab`,
`cell_type_popv`, `cell_type_biomni_run_{1,2,3}`, `cell_type_spatialagent_run_4`.

## Loading

```python
import anndata as ad
adata = ad.read_h5ad("annotation/combined_annotations_anonymized.h5ad")
adata.obs["cell_type_tier1_expert5"]          # one expert's tier-1 cell types
adata.layers["raw_count"]                      # raw counts
```

## Caveats
- **Expert 1** did not perform tissue-niche annotation (niche fields are empty/NA).
- **Expert 3**'s labels are of uncertain origin and likely mis-ordered — use with care.
- **Expert 8** has no standalone raw file; their annotations live only in the combined objects.
- `per_expert_raw/` files keep each expert's **native, heterogeneous** column names
  (only the filename was anonymized; no scientist name appears in any column or value).