Accurate and interpretable enhancement for single-cell chromatin accessibility sequencing data with scCASE

Installation

Environment and system requirements

We tested scCASE with Python 3.8, 3.9 and 3.10 on Ubuntu 22.04 LTS. We recommend to use Anoconda to setup the environment and manage dependencies.

User installation

scCASE is available on PyPI and can be installed via

pip install scCASE

You can also install scCASE from GitHub via

git clone git://github.com/BioX-NKU/scCASE.git
cd scCASE
python setup.py install

The dependencies will be automatically installed along with scCASE. Normally, the installation time does not exceed one minute.

Quick Start

Input format

• h5ad file:

  • AnnData object of shape n_obs × n_vars.

• count matrix file:

  • Rows correspond to peaks and columns to cells, in txt/tsv (sep="\t") or csv (sep=",") format.

Output format

• Enhanced scCAS data: The data enhanced by scCASE.
• Optional output:
  • Projection matrix(W): Projection matrix created by scCASE, which is the peak expression program.
  • Cell embedding matrix(H): Cell embedding created by scCASE, which is the low-dimensional representation of cells.
  • Similarity matrix(Z): Similarity matrix created by scCASE, which is the cell-to-cell similarity calculated through iteration.

Run scCASE

• Run scCASE to enhance scCASE data: import scCASE data_enhanced = scCASE.run(adata,method= "scCASE")

• Use scCASER to enhance scCASE data with reference data: data_enhanced = scCASE.run(data_path,ref_path,method= "scCASER")

• Use scCASE to correct sequencing depth: data_enhanced = scCASE.run(data_path,method= "Correct seq depth")

• Use scCASE to correct batch effect: data_enhanced = scCASE.run(data_path,method = "Correct batch effect",batchlabel = "batch")

Tutorials

Usage and examples of scCASE's main functions are showed in tutorial.

We provide a tutorial for running scCASE and integrating EpiScanpy for scCAS data analysis. The expected run time for the tutorial is less than 5 minutes.

The extension of scCASE can additionally correct sequencing depth, and we provide a tutorial.

To enhance datasets with batch effects, refer to the tutorial.

API

Main functions of scCASE:

run(data_path,ref_path = None,method = "scCASE",data_format = "h5ad", data_sep=",",ref_sep=",", K = "Auto",K_ref = "Auto", type_number_range=range(3, 15), save_result = False, output_path = "./", batchlabel= "batch",threshold = 0.01)

Function "scCASE.run()"is the main function of scCASE.

• [data_path] :str or ad.AnnData The path or variable name of the target scCAS data.
• [ref_path] ：str The path of the reference data,file format should be "csv".
• [method] ：str Enhancement methods including "scCASE", "scCASER", "Correct seq depth" or "Correct batch effect".
• [data_format] ：str The format of data and reference data, including "count matrix"(csv/txt/tsv) and "h5ad". Default: "h5ad"
• [data_sep] ：str The Separator of target scCAS data, only for count matrix file. Default: ","
• [ref_sep] : str The Separator of reference data, only for count matrix file. Default: ","
• [ K ] : "Auto" or int The parameter K. Default: "Auto"
• [K_ref] : "Auto" or int The parameter K_ref. Default: "Auto"
• [type_number_range] : range The range of estimate parameter K, only for K = "Auto". Default: range(3, 15)
• [save_result] : bool If save the enhanced result or not .Default: Flase
• [output_path] : str The path of the output, only for save_result is True. Default: "./"
• [batchlabel] : str The key of the observation stored batch label, only for method="Correct batch effect".
• [threshold] : float The threshold of filtering peaks. Default: 0.01

lazy(data)

Function "scCASE.lazy()" including TF-IDF, PCA, creating neighbors graph,and t-SNE/UMAP.

identify_specific_peaks(data_enhanced, type_ ,obs = "cell_type",peak_name = "peak")

Function "scCASE.identify_specific_peaks()" is used to identify specific peaks of given cell labels.

• [data_enhanced] : str The data enhanced by scCASE.
• [obs] : str The key of the observation to consider.
• [type_] : str The type in adata.obs to consider.
• [peak_name] : str The key of the variables annotation which represent peak names.

Reproduction

The source codes for the reproduction of results, all dependencies including version numbers, and datasets are publicly available at Zenodo.