Analysis of Stereo-seq mouse olfactory bulb dataset

In this tutorial, we demonstrate SpaRCL on the analysis of Stereo-seq mouse olfactory bulb dataset including

• Spatial reconstruction

• Mini-batch relational contrastive learning

• Spatial domain identification

• Finding differentially expressed genes

• Spatial domain annotation

The dataset is available at JinmiaoChenLab GitHub (data >> Stero-seq.tar.gz).

import numpy as np
import pandas as pd
import scanpy as sc
import matplotlib.pyplot as plt
import seaborn as sns

import SpaRCL as rcl

Data loading and preprocessing

We load the dataset and find top 8000 highly variable genes.

adata = sc.read_h5ad('./data/Stereo_seq.h5ad')
adata

AnnData object with n_obs × n_vars = 19527 × 27106
    obsm: 'spatial'

sc.pp.highly_variable_genes(adata, n_top_genes=8000, flavor='seurat_v3')

Spatial reconstruction

We perform spatial reconstruction to aggregate expression from spatial neighbors.

rcl.spatial_reconstruction(adata, n_neighbors=30)

Mini-batch relational contrastive learning

We select 4000 reference centers and then perform mini-batch relational contrastive learning on the reconstructed data.

sc.pp.pca(adata)
rcl.reference_centers(adata, n_centers=4000)

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\cluster\_kmeans.py:887: UserWarning: MiniBatchKMeans is known to have a memory leak on Windows with MKL, when there are less chunks than available threads. You can prevent it by setting batch_size >= 4096 or by setting the environment variable OMP_NUM_THREADS=1
  warnings.warn(

Since this step takes some time, we load previously computed AnnData.

# rcl.run_RCL_minibatch(adata)

adata = sc.read_h5ad('./data/Stereo_seq_results.h5ad')
adata

AnnData object with n_obs × n_vars = 19527 × 27106
    obs: 'reference_centers'
    var: 'highly_variable', 'highly_variable_rank', 'means', 'variances', 'variances_norm'
    uns: 'hvg', 'relation', 'pca', 'reference_centers', 'spatial_reconstruction'
    obsm: 'X_pca', 'relation', 'spatial'
    varm: 'PCs'
    layers: 'counts', 'log1p'

Spatial domain identification

We create a temporary AnnData using the sample-by-reference relation matrix and then interoperate with SCANPY to identify spatial domains.

adata_V = sc.AnnData(adata.obsm['relation'])
adata_V.obs_names = adata.obs_names
adata_V.obsm['spatial'] = adata.obsm['spatial']
adata_V

AnnData object with n_obs × n_vars = 19527 × 3983
    obsm: 'spatial'

sc.pp.pca(adata_V)
sc.pp.neighbors(adata_V)

sc.tl.louvain(adata_V, resolution=0.4)

adata.obs['louvain'] = adata_V.obs['louvain']

We exclude domain 5 from downstream analysis, since its average count is lower than 100.

sns.barplot(y=np.sum(adata.layers['counts'].toarray(), axis=1), x=adata.obs['louvain'])

<AxesSubplot:xlabel='louvain'>

adata_subset = adata[~adata.obs['louvain'].isin(['5']),:]

fig, axs = plt.subplots(figsize=(8, 7))

sc.pl.embedding(
    adata_subset,
    basis='spatial',
    color='louvain',
    size=50,
    palette=sc.pl.palettes.default_102,
    legend_loc='right margin',
    show=False,
    ax=axs,
)

plt.tight_layout()

Trying to set attribute `.uns` of view, copying.

Finding differentially expressed genes

We find the differentially expressed (DE) genes across identified domains and show the domains and their DE gene expression patterns in spatial coordinates.

sc.tl.rank_genes_groups(adata_subset, groupby='louvain', use_raw=False, layer='counts', method='t-test_overestim_var')

de_genes = pd.DataFrame(adata_subset.uns['rank_genes_groups']['names']).iloc[:10,:]
de_genes

	0	1	2	3	4	6	7
0	Pcp4	Vip	Cmss1	Apod	Nme7	Hbb-bs	Plp1
1	Nrxn3	Lcat	Gm42418	Ptn	Snap25	Hba-a1	Mbp
2	Ppp3ca	Fam98c	Cck	Ptgds	Scg2	Hbb-bt	Fth1
3	Kcnb2	Gm10610	Cdk8	Atp1a2	Gabra1	Hba-a2	Trf
4	Camk2n1	Ly6c1	Nrsn1	Apoe	Gm42418	Ptgds	Cldn11
5	Tshz1	Gm48843	Calb2	Fabp7	Cpe	Apex2	Mal
6	Meis2	Vmn1r67	Nxph1	S100a5	Uchl1	Mgp	Sox11
7	Pbx3	Esp34	Olfm1	Kctd12	Syt1	Igf2	Mag
8	Camk2b	Tnfsf15	Sparcl1	Gng13	Cltc	Dcn	Zfp704
9	Calm2	AC124606.2	Ptprd	Npy	Slc17a7	Acta2	Mobp

fig, axs = plt.subplots(figsize=(8, 7))

sc.pl.embedding(
    adata_subset,
    basis='spatial',
    color='louvain',
    groups='0',
    layer='log1p',
    size=50,
    palette=sc.pl.palettes.default_102,
    legend_loc='right margin',
    show=False,
    ax=axs,
)

plt.tight_layout()

fig, axs = plt.subplots(figsize=(8, 7))

sc.pl.embedding(
    adata_subset,
    basis='spatial',
    color=de_genes.iloc[0,0],
    layer='log1p',
    size=50,
    color_map='Reds',
    vmax='p99.9',
    show=False,
    ax=axs,
)

plt.tight_layout()

fig, axs = plt.subplots(figsize=(8, 7))

sc.pl.embedding(
    adata_subset,
    basis='spatial',
    color='louvain',
    groups='3',
    layer='log1p',
    size=50,
    palette=sc.pl.palettes.default_102,
    legend_loc='right margin',
    show=False,
    ax=axs,
)

plt.tight_layout()

fig, axs = plt.subplots(figsize=(8, 7))

sc.pl.embedding(
    adata_subset,
    basis='spatial',
    color=de_genes.iloc[0,3],
    layer='log1p',
    size=50,
    color_map='Reds',
    vmax='p99.9',
    show=False,
    ax=axs,
)

plt.tight_layout()

Spatial domain annotation

We annotate the spatial domains based on the anatomical annotation from Fu, H. et al. and marker genes.

map_dict = {
    '0': 'GCL',
    '1': 'EPL',
    '2': 'GL',
    '3': 'ONL',
    '4': 'MCL',
    '6': 'BLD',
    '7': 'RMS',
}

adata_subset.obs['annotation'] = adata_subset.obs['louvain'].map(map_dict).astype('category')

fig, axs = plt.subplots(figsize=(8.3, 7))

sc.pl.embedding(
    adata_subset,
    basis='spatial',
    color='annotation',
    size=50,
    palette=sc.pl.palettes.default_102,
    legend_loc='right margin',
    show=False,
    ax=axs,
)

plt.tight_layout()