GeneClust 0.0.1

Cofunctional grouping-based feature gene selection for unsupervised scRNA-seq clustering

GeneClust: cofunctional grouping-based feature gene selection for unsupervised scRNA-seq clustering

GeneClust is a computational feature selection method for scRNA-seq cell clustering. GeneClust groups genes into clusters from which genes are evaluated and selected with the aim of maximizing relevance, minimizing redundancy and preserving complementarity.

Dependencies

• numpy>=1.21.5
• pandas>=1.4.2
• anndata>=0.8.0
• setuptools>=59.5.0
• loguru>=0.6.0
• sklearn>=0.0
• scikit-learn>=1.1.1
• scanpy>=1.9.1
• scipy>=1.7.3
• leidenalg>=0.8.9

Installation

1. PyPI

You can directly install the package from PyPI.

2. Github

Also, You can download the package from Github and install it locally:

git clone https://github.com/ToryDeng/scGeneClust.git
cd scGeneClust/
python3 setup.py install --user

Two Versions of GeneClust

Version	Usage Scenarios
GeneClust-ps	1. Number of cells is small (e.g., several thousand) 2. Cell clustering performance is more important
GeneClust-fast	1. Number of cells is large (e.g., over 50,000) 2. Computational efficiency is more important

Example Code

from scGeneClust.utils import load_PBMC3k
from scGeneClust import scGeneClust

# load the PBMC3k dataset
raw_adata = load_PBMC3k()
# GeneClust-fast
selected_genes = scGeneClust(raw_adata, version='fast', n_gene_clusters=200, random_stat=2022, verbosity=2)
# GeneClust-ps
selected_genes = scGeneClust(raw_adata, version='ps', n_cell_clusters=7, scale=1000, top_percent_relevance=5, random_stat=2022, verbosity=2)

GeneClust expects raw counts. The output is an ndarray of selected features, which can be used in the downstream cell clustering analysis.