%matplotlib inline

Analyse your spatial data using sliding windows

This example shows how to use squidpy.tl.sliding_window() to divide the obs of an anndata.AnnData object into adjecent, potentially overlapping, windows.

import matplotlib.pyplot as plt

import squidpy as sq

First, let’s download the MIBI-TOF dataset.

adata = sq.datasets.mibitof()

This data set contains a cell type annotation in anndata.AnnData.obs\ ["Cluster"] and a slide annotation in anndata.AnnData.obs\ ["library_id"]

adata.obs

	row_num	point	cell_id	X1	center_rowcoord	center_colcoord	cell_size	category	donor	Cluster	batch	library_id
3034-0	3086	23	2	60316.0	269.0	7.0	408.0	carcinoma	21d7	Epithelial	0	point23
3035-0	3087	23	3	60317.0	294.0	6.0	408.0	carcinoma	21d7	Epithelial	0	point23
3036-0	3088	23	4	60318.0	338.0	4.0	304.0	carcinoma	21d7	Imm_other	0	point23
3037-0	3089	23	6	60320.0	372.0	6.0	219.0	carcinoma	21d7	Myeloid_CD11c	0	point23
3038-0	3090	23	8	60322.0	417.0	5.0	303.0	carcinoma	21d7	Myeloid_CD11c	0	point23
...	...	...	...	...	...	...	...	...	...	...	...	...
47342-2	48953	16	1103	2779.0	143.0	1016.0	283.0	carcinoma	90de	Fibroblast	2	point16
47343-2	48954	16	1104	2780.0	814.0	1017.0	147.0	carcinoma	90de	Fibroblast	2	point16
47344-2	48955	16	1105	2781.0	874.0	1018.0	142.0	carcinoma	90de	Imm_other	2	point16
47345-2	48956	16	1106	2782.0	257.0	1019.0	108.0	carcinoma	90de	Fibroblast	2	point16
47346-2	48957	16	1107	2783.0	533.0	1019.0	111.0	carcinoma	90de	Fibroblast	2	point16

3309 rows × 12 columns

Stratified by library, we now want to assign each cell to a sliding window of a given size.

sq.tl.sliding_window(
    adata=adata,
    library_key="library_id",  # to stratify by sample
    window_size=300,
    overlap=0,
    copy=False,  # we modify in place
)

Let’s inspect the column that the function has added to our data.

adata.obs["sliding_window_assignment"]

3034-0      point23_window_0
3035-0      point23_window_0
3036-0      point23_window_1
3037-0      point23_window_1
3038-0      point23_window_1
                 ...        
47342-2    point16_window_12
47343-2    point16_window_14
47344-2    point16_window_14
47345-2    point16_window_12
47346-2    point16_window_13
Name: sliding_window_assignment, Length: 3309, dtype: category
Categories (48, object): ['point23_window_0' < 'point8_window_0' < 'point16_window_0' < 'point23_window_1' ... 'point16_window_14' < 'point23_window_15' < 'point8_window_15' < 'point16_window_15']

We see that each observation has been assigned to a window, which is defined by the sliding_window_assignment column. We can visualise this using squidpy.pl.spatial_scatter().

sq.pl.spatial_scatter(
    adata, color="sliding_window_assignment", library_key="library_id", figsize=(10, 10)
)

Optionally, we can also look at a specific sample:

sq.pl.spatial_scatter(
    adata,
    color="sliding_window_assignment",
    library_key="library_id",
    library_id=["point8"],
    figsize=(10, 10),
)

We see that the function has created 16 windows, this is based on the window_size of 200 and an overlap of 0. The behaviour of the function changes when we use an overlap, since then observations will be assigned to multiple windows. This information can no longer be stored in a single column. Let’s try this out.

adata = sq.datasets.mibitof()  # fresh copy

sq.tl.sliding_window(
    adata=adata,
    library_key="library_id",  # to stratify by sample
    window_size=300,
    overlap=50,
    copy=False,  # we modify in place
)

When now inspecting the anndata.AnnData.obs, we see that several columns have been added, each indicating whether an observation is a member of a specific window, stratified by library_key.

Due to the overlapping assignments, we now have more “true” assignments than observations. This is because each observation can be a member of multiple windows.

total_cells = 0
for lib_key in ["point8", "point16", "point23"]:
    cols_in_lib = adata.obs.columns[adata.obs.columns.str.contains(lib_key)]
    total_cells_in_lib = sum(adata.obs[col].sum() for col in cols_in_lib)
    total_cells += total_cells_in_lib
    print(f"Total cells in {lib_key}: {total_cells_in_lib}")

print(f"Total cells: {total_cells}")

Total cells in point8: 1421
Total cells in point16: 1421
Total cells in point23: 1727
Total cells: 4569

We can also illustrate these overlapping windows.

adata_point8 = adata[adata.obs["library_id"] == "point8"]
cols = adata.obs.columns[adata.obs.columns.str.contains("point8")]

# convert True/False to category so we can vizualize it
adata_point8.obs[cols] = adata_point8.obs[cols].astype("category")

sq.pl.spatial_scatter(
    adata_point8,
    color=cols,
    library_key="library_id",
    library_id="point8",
    legend_loc=None,
)
plt.tight_layout()

Finally, we see that these specific parameters result in tiny windows with very few cells at the bottom and right corner. We can drop these with the parameter drop_partial_windows.

adata = sq.datasets.mibitof()  # fresh copy

sq.tl.sliding_window(
    adata=adata,
    library_key="library_id",  # to stratify by sample
    window_size=300,
    overlap=50,
    copy=False,  # we modify in place
    drop_partial_windows=True,
)

adata_point8 = adata[adata.obs["library_id"] == "point8"]
cols = adata.obs.columns[adata.obs.columns.str.contains("point8")]

# convert True/False to category so we can vizualize it
adata_point8.obs[cols] = adata_point8.obs[cols].astype("category")

sq.pl.spatial_scatter(
    adata_point8,
    color=cols,
    library_key="library_id",
    library_id="point8",
    legend_loc=None,
)
plt.tight_layout()

If desired, in-place modifications can be avoided by using copy=True. This then returns a pandas.DataFrame with the assignments.

adata = sq.datasets.mibitof()  # fresh copy

assignment = sq.tl.sliding_window(
    adata=adata,
    library_key="library_id",  # to stratify by sample
    window_size=300,
    overlap=50,
    copy=True,  # we modify in place
    drop_partial_windows=True,
)

assignment

	sliding_window_assignment_point23_window_0	sliding_window_assignment_point23_window_1	sliding_window_assignment_point23_window_2	sliding_window_assignment_point23_window_3	sliding_window_assignment_point23_window_4	sliding_window_assignment_point23_window_5	sliding_window_assignment_point23_window_6	sliding_window_assignment_point23_window_7	sliding_window_assignment_point23_window_8	sliding_window_assignment_point8_window_0	...	sliding_window_assignment_point16_window_1	sliding_window_assignment_point16_window_2	sliding_window_assignment_point16_window_3	sliding_window_assignment_point16_window_4	sliding_window_assignment_point16_window_5	sliding_window_assignment_point16_window_6	sliding_window_assignment_point16_window_7	sliding_window_assignment_point16_window_8	globalX	globalY
3034-0	True	False	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	7.0	269.0
3035-0	True	True	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	6.0	294.0
3036-0	False	True	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	4.0	338.0
3037-0	False	True	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	6.0	372.0
3038-0	False	True	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	5.0	417.0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
47342-2	False	False	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	1016.0	143.0
47343-2	False	False	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	1017.0	814.0
47344-2	False	False	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	1018.0	874.0
47345-2	False	False	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	1019.0	257.0
47346-2	False	False	False	False	False	False	False	False	False	False	...	False	False	False	False	False	False	False	False	1019.0	533.0

3309 rows × 29 columns

For reproducibility

import spatialdata

import numpy
import pandas

import matplotlib

import scanpy
import squidpy

%load_ext watermark

%watermark -v -m -p numpy,pandas,matplotlib,scanpy,squidpy,spatialdata

Python implementation: CPython
Python version       : 3.10.12
IPython version      : 8.3.0

numpy      : 1.23.4
pandas     : 2.2.2
matplotlib : 3.9.2
scanpy     : 1.10.2
squidpy    : 1.6.2.dev34+gb4a49c9.d20241030
spatialdata: 0.2.2

Compiler    : Clang 15.0.7 
OS          : Darwin
Release     : 22.2.0
Machine     : arm64
Processor   : arm
CPU cores   : 8
Architecture: 64bit