GiniClust3

GiniClust3: a fast and memory-efficient tool for rare cell type identification

GiniClust is a clustering method specifically designed for rare cell type detection. It uses the Gini index to identify genes that are associated with rare cell types without prior knowledge. This differs from traditional clustering methods using highly variable genes. Using a cluster-aware, weighted consensus clustering approach, we can combine the outcomes from Gini index and Fano factor-based clustering and identify both common and rare cell types. In this new version (GiniClust3), we have substantially increased the speed and reduced memory usage in order to meet the need for large data size. It can now be used to identify rare cell types from over a million cells. Previous versions of GiniClust can be found below: GiniClust (https://github.com/lanjiangboston/GiniClust). GiniClust2 (https://github.com/dtsoucas/GiniClust2).

GiniClust3 documentation is available through https://giniclust3.readthedocs.io/en/latest/, including installation instructions and tutorial.

If you use GiniClust v1.0-v3.0, please consider cite one or more of the following papers:

• Jiang L, Chen H, Pinello L, Yuan GC. GiniClust: detecting rare cell types from single-cell gene expression data with Gini index. Genome Biol. 2016 Jul 1;17(1):144. PMCID:PMC4930624
• Tsoucas D, Yuan GC. GiniClust2: a cluster-aware, weighted ensemble clustering method for cell-type detection. Genome Biol. 2018 May 10;19(1):58. PMCID:PMC5946416
• Dong R, Yuan GC. GiniClust3: a fast and memory-efficient tool for rare cell type identification. BMC Bioinformatics. 2020 Apr 25;21(1):158. PMCID:PMC7183612

A schematic overview of the GiniClust3 pipeline

<img src="https://github.com/rdong08/GiniClust3/blob/master/docs/images/pipeline.png" width="500">

Prerequisites

• numpy (https://numpy.org/)
• scikit-learn (https://scikit-learn.org/stable/)
• scipy (https://www.scipy.org/)
• statsmodels (https://www.statsmodels.org/stable/index.html)
• scanpy (https://scanpy.readthedocs.io/en/stable/)

Installation

Scanpy is needed to be installed first from "https://scanpy.readthedocs.io/en/stable/installation.html".

Install by using anaconda (recommend)

conda install -c rdong giniclust3

OR download from Github and install

python setup.py install

OR install by using pip

pip install giniclust3

Usage and example:

Import associated packages

import scanpy as sc
import numpy as np
import giniclust3 as gc
import anndata

Read single cell file

adataRaw=sc.read_csv("./data/GSM1599495_ES_d0_biorep_techrep1.csv",first_column_names=True)

Filter expression matrix

sc.pp.filter_cells(adataRaw,min_genes=3)
sc.pp.filter_genes(adataRaw,min_cells=200)

Format expression matrix

###example csv file is col:cells X row:genes. Skip this step if the input matrix is col:genes X row:cells
adataSC=anndata.AnnData(X=adataRaw.X.T,obs=adataRaw.var,var=adataRaw.obs)

Normalization

sc.pp.normalize_per_cell(adataSC, counts_per_cell_after=1e4)

Perform GiniIndexClust

gc.gini.calGini(adataSC) ###Calculate Gini Index
adataGini=gc.gini.clusterGini(adataSC,neighbors=3) ###Cluster based on Gini Index

Perform FanoFactorClust

gc.fano.calFano(adataSC) ###Calculate Fano factor
adataFano=gc.fano.clusterFano(adataSC) ###Cluster based on Fano factor

ConsensusClust

consensusCluster={}
consensusCluster['giniCluster']=np.array(adataSC.obs['rare'].values.tolist())
consensusCluster['fanoCluster']=np.array(adataSC.obs['fano'].values.tolist())
gc.consensus.generateMtilde(consensusCluster) ###Generate consensus matrix
gc.consensus.clusterMtilde(consensusCluster) ###Cluster consensus matrix
np.savetxt("final.txt",consensusCluster['finalCluster'], delimiter="\t",fmt='%s')

UMAP visualization

adataGini.obs['final']=consensusCluster['finalCluster']
adataFano.obs['final']=consensusCluster['finalCluster']
gc.plot.plotGini(adataGini)
gc.plot.plotFano(adataFano)

Citation

Dong R, Yuan GC. GiniClust3: a fast and memory-efficient tool for rare cell type identification. BMC Bioinformatics. 2020 Apr 25;21(1):158. PMCID:PMC7183612

License

Copyright (C) 2019 YuanLab. See the LICENSE file for license rights and limitations (MIT).