scLKME

A Landmark-based Approach for Generating Multi-cellular Sample Embeddings from Single-cell Data

Introduction

scLKME is a computational approach designed for generating sample embeddings in multi-sample single-cell data. scLKME models samples as cell distributions, and leverages lanbmark-based kernel mean embedding to generate a sample-by-landmark kernel matrix as sample embeddings. The workflow of scLKME is as follows:

Installation

scLKME requires python version >= 3.8, < 3.12.

scLKME including its dependencies can be installed via PyPI by running:

pip install sclkme

Data

All the datasets used in this study are shared via .h5ad format. The datasets are available at:

• Multi-species: here
• Preterm: here
• Myocardial Infarction: here

Get started

scLKME's apis are designed to be compatible with the scanpy's ecosystem. To use scLKME, here is a basic example:

import sclkme
import scanpy as sc

n_sketch= 128
sample_id = "sample_id"

adata = sc.read_h5ad("path_to_h5ad_data")
sclkme.tl.sketch(adata, n_sketch=n_sketch)

X_anchor = adata[adata.obs['sketch']].X.copy()
sclkme.tl.kernel_mean_embedding(adata, partition_key="sample_id", X_anchor=X_anchor)

# sample_embedding
adata.uns['kme'][f'{sample_id}_kme']

For more details, examples and tutorials, check our document.

Run on a cloud platform

Tutorials	Colab
Cell Sketching
Landmark-based multi-sample single-cell data analysis

Citation

@article{yi2023sclkme,
  title={scLKME: A Landmark-based Approach for Generating Multi-cellular Sample Embeddings from Single-cell Data},
  author={Yi, Haidong and Stanley, Natalie},
  journal={bioRxiv},
  pages={2023--11},
  year={2023},
  publisher={Cold Spring Harbor Laboratory}
}