Cluster neighborhoods and embeddings

To further investigate the neuronal diversity within each major brain structure, Yao et al created concept of cell types “neighborhoods” or groups and use them for visualization and analysis. For each neighborhood, they genereated re-embedded UMAPS to reveal fine-grained relationships between neuronal types within and between brain regions.

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from pathlib import Path

from abc_atlas_access.abc_atlas_cache.abc_project_cache import AbcProjectCache

We will interact with the data using the AbcProjectCache. This cache object tracks which data has been downloaded and serves the path to the requsted data on disk. For metadata, the cache can also directly serve a up a Pandas Dataframe. See the getting_started notebook for more details on using the cache including installing it if it has not already been.

Change the download_base variable to where you have downloaded the data in your system.

download_base = Path('../../data/abc_atlas')
abc_cache = AbcProjectCache.from_cache_dir(download_base)

abc_cache.current_manifest

'releases/20241130/manifest.json'

Data Overview

Cluster groups

Each cell type neighborhood is represented as a cluster group. Each row of the dataframe represnts a cluster group. Each cluster has a label (human readable string that is unique in the database), name, description and the number of clusters that has been grouped together.

groups = abc_cache.get_metadata_dataframe(directory='WMB-neighborhoods', file_name='cluster_group')
print("number of groups:",len(groups))
groups.set_index('label', inplace=True)
groups

cluster_group.csv: 100%|████████████████████████████████████████████████████████████████████████████████| 1.62k/1.62k [00:00<00:00, 28.8kMB/s]

number of groups: 8

	name	description	group_type	number_of_clusters
label
CCN20230722_HY-EA-Glut-GABA	HY-EA-Glut-GABA	Glutamatergic and GABAergic neurons in hypotha...	cell type neighborhood	1404
CCN20230722_MB-HB-CB-GABA	MB-HB-CB-GABA	GABAergic neurons in midbrain, hindbrain, and ...	cell type neighborhood	1040
CCN20230722_MB-HB-Glut-Sero-Dopa	MB-HB-Glut-Sero-Dopa	Glutamatergic, serotonergic and dopaminergic n...	cell type neighborhood	1431
CCN20230722_NN-IMN-GC	NN-IMN-GC	Non-neuronal cells, immature neurons, granule ...	cell type neighborhood	248
CCN20230722_Pallium-Glut	Pallium-Glut	Glutamatergic neurons from cerebral cortex (de...	cell type neighborhood	517
CCN20230722_Subpallium-GABA	Subpallium-GABA	Telencephalic GABAergic neurons from cerebral ...	cell type neighborhood	1051
CCN20230722_TH-EPI-Glut	TH-EPI-Glut	Glutamatergic neurons from thalamus and epitha...	cell type neighborhood	148
CCN20230722_WholeBrain	WholeBrain	All cells in the whole mouse brain (5322 clust...	cell type neighborhood	5322

# print out description for each group
for tsindex, tsrow in groups.iterrows() :
    print("%s:\n" % tsrow['name'])
    print("%s\n" % tsrow['description'])

HY-EA-Glut-GABA:

Glutamatergic and GABAergic neurons in hypothalamus and extended amygdala, also including all non-pallium glutamatergic neurons in cerebral nuclei (1404 clusters)

MB-HB-CB-GABA:

GABAergic neurons in midbrain, hindbrain, and cerebellum (1040 clusters)

MB-HB-Glut-Sero-Dopa:

Glutamatergic, serotonergic and dopaminergic neurons in midbrain and hindbrain (1431 clusters)

NN-IMN-GC:

Non-neuronal cells, immature neurons, granule cells in dentate gyrus, olfactory bulb and cerebellum, and neurons in pineal gland (248 clusters)

Pallium-Glut:

Glutamatergic neurons from cerebral cortex (derived from the developmental pallium), including isocortex, hippocampal formation, olfactory areas, and cortical subplate (517 clusters)

Subpallium-GABA:

Telencephalic GABAergic neurons from cerebral cortex and cerebral nuclei, as well as some from the hypothalamic preoptic area, which are derived from the developmental subpallium (1051 clusters)

TH-EPI-Glut:

Glutamatergic neurons from thalamus and epithalamus (i.e., medial and lateral habenula) (148 clusters)

WholeBrain:

All cells in the whole mouse brain (5322 clusters)

Cluster group membership

The association between a cluster and cluster group is represented as a cluster group membership.

group_membership = abc_cache.get_metadata_dataframe(directory='WMB-neighborhoods', file_name='cluster_group_membership')
group_membership.head(5)

cluster_group_membership.csv: 100%|███████████████████████████████████████████████████████████████████████| 674k/674k [00:00<00:00, 2.49MMB/s]

	cluster_label	cluster_group_label	cluster_alias	cluster_group_name
0	CS20230722_0128	CCN20230722_Pallium-Glut	128	Pallium-Glut
1	CS20230722_0128	CCN20230722_WholeBrain	128	WholeBrain
2	CS20230722_0129	CCN20230722_Pallium-Glut	129	Pallium-Glut
3	CS20230722_0129	CCN20230722_WholeBrain	129	WholeBrain
4	CS20230722_0130	CCN20230722_Pallium-Glut	130	Pallium-Glut

Dimension reductions

Principal component analysis (PCA) was first preformed on the gene expression matrix. The top 100 PCs were selected and used to create a 2D UMAP (Uniform Manifold Approximation and Projection for Dimension Reduction) for all cells in the datasets. Further, for each neighborhood or cluster group, an additional specific 2D UMAP or embedding is created.

Each row of the dimension reduction dataframe represents on embedding. Each row of the dataframe represnts an embedding. Each dimension reduction has a label (human readable string that is unique in the database), name, description, the number of cells in the embedding and the label of the associated cluster group.

maps = abc_cache.get_metadata_dataframe(directory='WMB-neighborhoods', file_name='dimension_reduction')
print("number of maps:", len(maps))
maps

dimension_reduction.csv: 100%|██████████████████████████████████████████████████████████████████████████| 1.68k/1.68k [00:00<00:00, 31.6kMB/s]

number of maps: 8

	label	name	description	number_of_cells	cluster_group_label
0	UMAP20230830-HY-EA-Glut-GABA	HY-EA-Glut-GABA	Glutamatergic and GABAergic neurons in hypotha...	295448	CCN20230722_HY-EA-Glut-GABA
1	UMAP20230830-MB-HB-CB-GABA	MB-HB-CB-GABA	GABAergic neurons in midbrain, hindbrain, and ...	186775	CCN20230722_MB-HB-CB-GABA
2	UMAP20230830-MB-HB-Glut-Sero-Dopa	MB-HB-Glut-Sero-Dopa	Glutamatergic, serotonergic and dopaminergic n...	180669	CCN20230722_MB-HB-Glut-Sero-Dopa
3	UMAP20230830-NN-IMN-GC	NN-IMN-GC	Non-neuronal cells, immature neurons, granule ...	1418529	CCN20230722_NN-IMN-GC
4	UMAP20230830-Pallium-Glut	Pallium-Glut	Glutamatergic neurons from cerebral cortex (de...	1494801	CCN20230722_Pallium-Glut
5	UMAP20230830-Subpallium-GABA	Subpallium-GABA	Telencephalic GABAergic neurons from cerebral ...	611423	CCN20230722_Subpallium-GABA
6	UMAP20230830-TH-EPI-Glut	TH-EPI-Glut	Glutamatergic neurons from thalamus and epitha...	126171	CCN20230722_TH-EPI-Glut
7	UMAP20230830-WholeBrain	WholeBrain	All cells in the whole mouse brain (5322 clust...	4042976	CCN20230722_WholeBrain

Example use cases

Cluster group membership pivot table

To support analysis and visualization by cluster group in later tutorials, we create a pivot table using pandas groupby fuunction. Each row of the resulting dataframe represents a cluster, each column represents cluster group and the value in the table is a boolean specifying if the cluster is a member of the cluster group.

pivot = group_membership.groupby(['cluster_alias','cluster_group_name'])['cluster_group_name'].first().unstack()
pivot = pivot.notna()
pivot

cluster_group_name	HY-EA-Glut-GABA	MB-HB-CB-GABA	MB-HB-Glut-Sero-Dopa	NN-IMN-GC	Pallium-Glut	Subpallium-GABA	TH-EPI-Glut	WholeBrain
cluster_alias
1	False	False	False	False	True	False	False	True
2	False	False	False	False	True	False	False	True
3	False	False	False	False	True	False	False	True
4	False	False	False	False	True	False	False	True
5	False	False	False	False	True	False	False	True
...	...	...	...	...	...	...	...	...
34368	False	True	False	False	False	False	False	True
34372	False	True	False	False	False	False	False	True
34374	False	True	False	False	False	False	False	True
34376	False	True	False	False	False	False	False	True
34380	False	True	False	False	False	False	False	True

5322 rows × 8 columns

The pivot dataframe created above is stored as part of the cache as cluster_group_membership_pivoted.

Visualize a cluster group in whole brain and neighborhood embeddings

In this example, we visualize cluster group “CCN20230722_HY-EA-Glut-GABA” in context the whole brain UMAP and specific “HY-EA-Glut-GABA” UMAP.

First step is to read in 10x cell metadata with annotation dataframe.

cell = abc_cache.get_metadata_dataframe(
    directory='WMB-10X',
    file_name='cell_metadata_with_cluster_annotation',
    dtype={"neurotransmitter": str}
)
cell.set_index('cell_label', inplace=True)

We join with cluster membership pivot table to get cell to cluster group member. For convenience, we can cache this view for later reuse.

joined_10x = cell.join(pivot, on='cluster_alias')

We defined a helper function to plot a UMAP using matplotlib scatter function.

def plot_umap(xx, yy, cc = None, val = None, fig_width = 8, fig_height = 8, cmap = None) :
    
    fig, ax = plt.subplots()
    fig.set_size_inches(fig_width, fig_height)
    
    if cmap is not None :
        plt.scatter(xx, yy, s=0.5, c=val, marker='.', cmap=cmap)
    elif cc is not None :
        plt.scatter(xx, yy, s=0.5, color=cc, marker='.')
        
    ax.axis('equal')
    ax.set_xlim(-18, 27)
    ax.set_ylim(-18, 27)
    ax.set_xticks([])
    ax.set_yticks([])
    
    return fig, ax

First we plot all cells in the whole brain UMAP colorized by cell type classes

subsampled = joined_10x.loc[::10]
color = subsampled['class_color'].copy()
fig, ax = plot_umap(subsampled['x'], subsampled['y'], cc=color)
res = ax.set_title('WholeBrain: all classes')
plt.show()

We filter to only colorized cells in the “HY-EA-Glut-GABA” cluster group to visualize cell type classes, subclasses and neurotransmitter types

glabel = "CCN20230722_HY-EA-Glut-GABA"
print(groups.loc[glabel, 'description'])
gname = groups.loc[glabel, 'name']
pred = (subsampled[gname] == False)

color = subsampled['class_color'].copy()
color[pred] = '#f5f5f5'
fig, ax = plot_umap(subsampled['x'], subsampled['y'], cc=color)
res = ax.set_title('WholeBrain: HY-EA classes')
plt.show()

Glutamatergic and GABAergic neurons in hypothalamus and extended amygdala, also including all non-pallium glutamatergic neurons in cerebral nuclei (1404 clusters)

color = subsampled['subclass_color'].copy()
color[pred] = '#f5f5f5'
fig, ax = plot_umap(subsampled['x'], subsampled['y'], cc=color)
res = ax.set_title('WholeBrain: HY-EA subclasses')
plt.show()

color = subsampled['neurotransmitter_color'].copy()
color[pred] = '#f5f5f5'
fig, ax = plot_umap(subsampled['x'], subsampled['y'], cc=color)
res = ax.set_title('WholeBrain: HY-EA subclasses')
plt.show()

In this next section, we read in the coordinates for the HY-EA-Glut-GABA specific embedding

maps_by_group = maps.set_index('cluster_group_label')
mlabel = maps_by_group.loc[glabel,'label']

coord = abc_cache.get_metadata_dataframe(directory='WMB-neighborhoods', file_name=mlabel)
coord.set_index('cell_label', inplace=True)

UMAP20230830-HY-EA-Glut-GABA.csv: 100%|█████████████████████████████████████████████████████████████████| 13.7M/13.7M [00:00<00:00, 19.9MMB/s]

We create a new dataframe “neighborhoos” by joining the spatial coordinates with cell metadata

neighborhood = coord.join(joined_10x, how='left', rsuffix='orig')

We can resuse the same plot umap helper function to visualize cell type classes, subclasses and neurotransmitter types in the new embedding

fig, ax = plot_umap(neighborhood['x'], neighborhood['y'], cc=neighborhood['class_color'])
res = ax.set_title('HY-EA-Glut-GABA classes')
plt.show()

fig, ax = plot_umap(neighborhood['x'], neighborhood['y'], cc=neighborhood['subclass_color'])
res = ax.set_title('HY-EA-Glut-GABA subclasses')
plt.show()

fig, ax = plot_umap(neighborhood['x'], neighborhood['y'], cc=neighborhood['neurotransmitter_color'])
res = ax.set_title('HY-EA-Glut-GABA neurotransmitter types')
plt.show()

Visualize gene expression in neighborhood embedings

In this section, we show how to visualize gene expression in a neighborhood specific embedding. We start by reading in the gene expression dataframe we previously cached in the metadata directory from running 10x part 2a tutorial.

exp = abc_cache.get_metadata_dataframe(directory='WMB-10X', file_name='example_genes_all_cells_expression')
exp.set_index('cell_label', inplace=True)
print(exp.columns)

Index(['Slc32a1', 'Slc17a7', 'Slc6a5', 'Slc17a6', 'Slc6a2', 'Slc17a8', 'Tac2',
       'Slc6a4', 'Slc6a3', 'Slc18a3'],
      dtype='object')

We create a new dataframe “neighborhood_exp” that joins the neighborhood with gene expression information. We resuse the plot umap helper function to plot expression for genes Slc32a1 (GABAergic enriched) and Slc17ab (Glutamergic enriched).

neighborhood_exp = neighborhood.join(exp)

gene = 'Slc32a1'
fig, ax = plot_umap(neighborhood_exp['x'], neighborhood_exp['y'], val=neighborhood_exp[gene], cmap=plt.cm.magma_r)
res = ax.set_title('HY-EA-Glut-GABA: %s' % gene)
plt.show()

gene = 'Slc17a6'
fig, ax = plot_umap(neighborhood_exp['x'], neighborhood_exp['y'], val=neighborhood_exp[gene], cmap=plt.cm.magma_r)
res = ax.set_title('HY-EA-Glut-GABA: %s' % gene)
plt.show()

Visualize cluster groups in anatomical context

In this section, we visualize a cluster group in anatomical context using the whole brain merfish dataset. We begin by reading in the extended cell metadata dataframe.

cell = abc_cache.get_metadata_dataframe(
    directory='MERFISH-C57BL6J-638850',
    file_name='cell_metadata_with_cluster_annotation',
    dtype={"cell_label": str}
)
cell.set_index('cell_label', inplace=True)

cell_metadata_with_cluster_annotation.csv: 100%|████████████████████████████████████████████████████████| 1.02G/1.02G [00:40<00:00, 25.1MMB/s]

We join with cluster membership pivot table to get cell to cluster group member. For convenience, we can cache this view for later reuse.

joined_merfish = cell.join(pivot, on='cluster_alias')

We create a helper function to visualized a set of brain sections, colorized by specified cell property and filtered to a user specified cluster group.

def plot_sections(df, group, feature, blist, fig_width = 20, fig_height = 5) :
    
    fig, ax = plt.subplots(1, len(blist))
    fig.set_size_inches(fig_width, fig_height)
    
    for idx,bsl in enumerate(blist):
        
        filtered = df[df['brain_section_label'] == bsl]
        xx = filtered['x']
        yy = filtered['y']

        vv = filtered[feature].copy()
        pred = (filtered[group] == False)
        vv[pred] = '#f5f5f5'
        
        ax[idx].scatter(xx, yy, s=1.0, color=vv, marker=".")
            
        ax[idx].axis('equal')
        ax[idx].set_xlim(0, 11)
        ax[idx].set_ylim(11, 0)
        ax[idx].set_xticks([])
        ax[idx].set_yticks([])
        
        ax[idx].set_title("%s: %s" % (group, bsl))

        
    plt.subplots_adjust(wspace=0.01, hspace=0.01)
    plt.show()
    return fig, ax

We use the helper function to view 6 MERFISH sections, colorized by cell type subclass and restricted to HY-EA-Glut-GABA subgroup.

blist = ['C57BL6J-638850.46', 'C57BL6J-638850.43', 'C57BL6J-638850.40']
fig, ax = plot_sections(joined_merfish, 'HY-EA-Glut-GABA', 'subclass_color', blist)
plt.show()

blist = ['C57BL6J-638850.38', 'C57BL6J-638850.36', 'C57BL6J-638850.32']
fig, ax = plot_sections(joined_merfish, 'HY-EA-Glut-GABA', 'subclass_color', blist)
plt.show()