PHD-MS: Persistent Homology for Domains at Multiple Scales

Multiscale domain identification for spatial transcriptomic data.

Installation

Simply install with pip:

pip install phd_ms

Or install from source by downloading, cd to this directory, using:

pip install .

Usage/Examples

Detailed jupyter notebook tutorials are available in the examples folder. Or simply download the file titled point_and_click.py and run it to use our clickable graphical interface. When using point_and_click.py, set the path to your data by changing this line:

DATA = '/path/to/your_data.h5ad'

Here we'll show a simple example with Visium DLPFC data, using default parameters. First, import necessary components.

import numpy as np
import phd_ms

Preprocessing steps here (note that we assume a spatially-aware embedding has already been computed for your data and stored in adata.obsm):

INPUT_FILE = '/path/to/adata_151673.h5ad'
RESOLUTIONS = np.linspace(start=0.15, stop=.95, num=8)
RES_KEYS = [f'leiden_{r:.3f}' for r in RESOLUTIONS]
adata = phd_ms.tl.preprocess_leiden(INPUT_FILE, output_file=INPUT_FILE, emb='X_gst',
                                    resolution=RESOLUTIONS, res_keys=RES_KEYS)

Compute persistent homology and plot the 10 most prominent multiscale domains:

cluster_complex, clusterings = phd_ms.tl.cluster_filtration(adata, res_keys=RES_KEYS)
domains = phd_ms.tl.map_multiscale(adata.obsm['spatial'], cluster_complex, clusterings, num_domains=10)