svgbit 0.2.17

Find spatial variable genes for Spatial Trasncriptomics data.

For further detail, please visit our API reference on readthedocs.org

Folder slides contains 10X Visium slide layout GPR files used in our publication. Please refer to our paper’s supplementary table 1 for sample index.

Installation

Install svgbit with pip:

pip install svgbit

Command-line Interface

svgbit has a command line version. Just tape:

svgbit --help

after installation, and you may get a short help massage:

usage: svgbit [-h] [--k K] [--n_svgs N_SVGS] [--n_svg_clusters N_SVG_CLUSTERS]
              [--he_image HE_IMAGE] [--save_dir SAVE_DIR] [--cores CORES] read_path

Find spatial variable genes for Spatial Trasncriptomics data.

positional arguments:
read_path             Read Spatial Transcriptomics data. Support format in 10X
                      Space Ranger(dir named ``outs``) and anndata hdf5


optional arguments:
-h, --help            show this help message and exit
--normalization NORMALIZATION
                      apply which normalization on read data. If None (default),
                      neither normalization will apply. Supported values: None, cpm,
                      logcpm (default: None)

--k K                 number of nearest neighbors for KNN network (default: 6)
--n_svgs N_SVGS       number of SVGs to find clusters (default: 1000)
--n_svg_clusters N_SVG_CLUSTERS
number of SVG clusters to find (default: 8)
--he_image HE_IMAGE   path to H&E image. Only used for visualization (default: None)
--save_dir SAVE_DIR   path to save results (default: .)
--cores CORES         number of threads to run svgbit (default: 8)

Follow the introduction and results will save to –savedir.

Note

svgbit will use all available CPUs as default. While python’s child process will inherit all resources from parent, this may consume much memory. Specify cores keyword argument to avoid memory error.

Note

svgbit may consume ~35 Gib memory when running with a 2980 spots, 32285 genes matrix with 8 cores.

Python Interface

svgbit has a set of python API. You may run svgbit through command line or python. We recommend the usage of python API for more feature and convient control of your input data.

Load Data

svgbit could load data from Space Ranger output directory:

import svgbit as sb
dataset = sb.load_10X("spaceranger_output/outs")

Or load data from csv files:

import svgbit as sb
dataset = sb.STDataset(
    count_df="Data/count_df.csv",
    coordinate_df="Data/coor_df.csv",
    count_df_kwargs={"index_col": 0, "header": 0},
    coordinate_df_kwargs={"index_col": 0, "header": 0},
)

Note

sb.STDataset also accept pd.DataFrame and np.array as count_df and coordinate_df. If str or pathlib.Path are given, svgbit would try to load data with pandas.

Note

When init STDataset instance in this way, svgbit assume that the shape of count matrix and coordinate file is (spot * gene) and (spot * 2). Specify count_transpose or coordinate_transpose as True when necessary.

Data Preprocessing

Genes with low variance may filter with:

dataset = sb.filters.low_variance_filter(dataset, var=0)

Note

With var=0, genes with no expression may be filtered. We recommend apply a var=0 filter first for better computational performing.

Genes with extremely high expressions usually show no pattern and may distrub performing. Filter with:

dataset = sb.filters.quantile_filter(dataset, 0.99)

svgbit alse has count normalization functions:

dataset = sb.normalizers.logcpm_normailzer(dataset)

Feel free for choosing gene filters and data normalizers. Other filters and normalizers are also provided. Visit our API reference for further detail.

Run svgbit

To perform full pipeline of svgbit, running:

sb.run(dataset)

Visualization

Draw SVG heatmap and spot type distribution map with:

sb.plot.svg_heatmap(dataset, save_path="heatmap.jpg", he_image="he_image.jpg")
sb.plot.spot_type_map(dataset, save_path="spot_type.jpg", he_image="he_image.jpg")

Parameter he_image is optional. If not specified, hotspot discription map will show without morphological information.

Gene combinations

Users may find gene combinations with:

sb.find_combinations(dataset, center_spots=1)

to find gene combinations for SVG cluster 1. A pd.DataFrame with gene colocalization and exclusive result will be returned.

Details about sb.run()

When you perform sb.run(), sevaral steps will be done as below. For further detail of calculation, please refer to our publication.

Acquire weight

To calculate hotspot matrix, svgbit needs a weight network which discribes association across spots. svgbit uses k-nearest neighbors with 6 neighbors as a default. You may pass key word argument k to sb.run() to change this behavior.

In this step, sb.run() will execute STDataset.acquire_weight() method with given parameters. You may also perform this step by:

dataset.acquire_weight()

Weight will save as attribute weight of STDataset and detailed discription of weight is saved to weight_type attribute. Users may provide a libpysal.weights.W instance as user-specified weight:

dataset.weight = user_specified_weight

Acquire hotspot

Hotspot matrix is estimated by:

dataset.acquire_hotspot()

and save to hotspot_df attribute.

Density

AI and Di value discribed in our paper will be calculate by:

dataset.acquire_density()

and save to AI and Di attribute as pd.Series.

Find SVG clusters

SVG clusters is estimated by:

dataset.find_clusters()

and save to svg_cluster attribute.

For further discription of hotspot, AI, Di and SVG cluster, please refer to our manuscript.

Citation

If you use SVGbit in your academic publication, please cite:

> Hong, Y., Song, K., Zhang, Z. et al. The spatiotemporal dynamics of spatially variable genes in developing mouse brain revealed by a novel computational scheme. Cell Death Discov. 9, 264 (2023). https://doi.org/10.1038/s41420-023-01569-w

For test codes used in our paper, please visit https://github.com/fuyu-sama/svgbit-test