| name | iblatlas-genomics-merfish |
| description | Load the MERFISH single-cell atlas (parquet dataframes) and the MERFISH cell-type density volumes using iblatlas, including how to denoise the volumes. Use when the user asks to load MERFISH, single-cell atlas data, cell-type data, MERFISH volumes, or how to denoise/preprocess them. |
| allowed-tools | ["Read"] |
name: ea-load-merfish-dataframes
Load MERFISH single-cell atlas
Read sources/examples/05_load_merfish.py for the canonical pattern and adapt it to context.
Outputs
| Variable | Description |
|---|
df_cells | One row per cell; columns x, y, z (MNI metres), class, subclass, supertype, cluster (integer keys) |
df_classes | Cell-class metadata; key column: class_rgba (packed int, convert with merfish.int2rgb()) |
df_subclasses | Subclass metadata |
df_supertypes | Supertype metadata |
df_clusters | Cluster metadata |
df_neurotransmitters | Neurotransmitter metadata |
Key points
- Cell coordinates are in metres (MNI space); multiply by
1e6 for µm when plotting.
- Get RGB colours for scatter plots:
rgb = merfish.int2rgb(df_classes.loc[df_cells['class'], 'class_rgba']).
- Filter a coronal slice:
mask = np.abs(df_cells['y'] - yslice) < (30 / 1e6).
- Use
df_cells['class'] as an index into df_classes (and similarly for other taxonomy levels).
Load MERFISH cell-type density volumes
from iblatlas.genomics import merfish
volume, labels, agea_atlas = merfish.load_volume(level='class')
| Variable | Shape | Description |
|---|
volume | (n_types, dim0, dim1, dim2) | Per-voxel cell-type density |
labels | (n_types,) | Type index per channel — same key space as df_classes.index above |
agea_atlas | — | Atlas matching the volume grid |
level is one of 'class', 'subclass', 'supertype', 'cluster'. The grid is the AGEA 200 µm
grid, not the default 25 µm AllenAtlas() — index it via agea_atlas.bc. See the
merfish.load_volume / merfish.denoise_volume docstrings for the label argument and denoising
details, and sources/examples/08_load_merfish_volumes.py for a worked loading + plotting example.
