cellqc: standardized quality control pipeline of single-cell RNA-Seq data

Cellqc standardizes the qualiy control of single-cell RNA-Seq (scRNA) data to render clean feature count matrices from Cell Ranger outputs. Cellqc is implemented using the Snakemake workflow management system to enhance reproduciblity and scalablity of data analysis. Briefly, the QC pipeline starts from raw count feature matrices from Cell Ranger. Dropkick filters out predicted empty droplets, and SoupX purify the transcriptome measurement by substracting the background trancripts. DoubletFinder further detects the potential doublets and retain clean count feature matrices for singlets. Cell types are annotated for clean cells by a reference database using scPred.

Installation

It is easy to install cellqc via conda at https://anaconda.org/bioconda/cellqc. To use the full function of cellqc, please also install several dependencies outside conda. It is encouraged to use the C++ implementation mamba to speed up the installation. E.g.,

conda config --add channels defaults --add channels bioconda --add channels conda-forge

# Downgrade Seurat to v4 for SeuratDisk, as Seurat v5 is not supported in SeuratDisk.
mamba create -y -n cellqc python=3.10 cellqc r-seurat=4 r-seuratobject=4 r-matrix=1.6.1 dropkick r-hdf5r hdf5 r-leidenbase libxml2 r-xml r-xml2 zlib bioconductor-rsamtools bioconductor-genomicfeatures bioconductor-rtracklayer 'pandas<2'

conda activate cellqc

# Build from source
Rscript -e "remotes::install_github(c('mojaveazure/seurat-disk', 'immunogenomics/harmony', 'powellgenomicslab/scPred', 'powellgenomicslab/DropletQC'), upgrade=F)"

# Bug fix @counts for Seurat object, instead of chris-mcginnis-ucsf/DoubletFinder
Rscript -e "remotes::install_github('lijinbio/DoubletFinder', upgrade=F, force=T)"

pip install -U cellqc # Optional: to install the latest version from PyPI

Dependent software are summarized below.

Software	URL
DoubletFinder	https://github.com/chris-mcginnis-ucsf/DoubletFinder
DropletUtils	https://bioconductor.org/packages/release/bioc/html/DropletUtils.html
Seurat	https://satijalab.org/seurat
SeuratDisk	https://github.com/mojaveazure/seurat-disk
SoupX	https://github.com/constantAmateur/SoupX
scPred	https://github.com/powellgenomicslab/scPred
DropletQC	https://github.com/powellgenomicslab/DropletQC
Snakemake	https://github.com/snakemake/snakemake
Scanpy	https://scanpy.readthedocs.io/en/stable
dropkick	https://github.com/KenLauLab/dropkick

To test the installation, simply run

cellqc -h

Run the pipeline

Cellqc requires a sample file for sample information and an optional configuration file for pipeline parameters.

• The sample file (e.g., samples.txt) is a tab-delimited file with headers: sample, cellranger, and/or nreaction.

  • The sample column is the sample ID per sample.
  • The cellranger is the Cell Ranger output directory. See Cell Ranger Outputs for an example directory.
  • The optional third column nreaction is the number of reactions in the library preparation, which is useful to infer expected doublets for a sample with a Cell Ranger analysis using combined raw reads from multiple reactions. If the nreaction column is not specified in the sample file, the default 1 reaction is used for all samples.

• A configuration file is in the YAML format. It is optional. The default parameters can be used as below. See the next section for the inspection of configuration.

nuclear_fraction
  numthreads: 12
  cbtag: CB
  retag: RE
dropkick:
  skip: false
  method: multiotsu
  numthreads: 1
filterbycount:
  mincount: 500
  minfeature: 300
  mito: 10
doubletfinder:
  skip: false
  findpK: false
  numthreads: 5
  pK: 0.01
scpred:
  skip: true
  reference: /path_to_reference/scPred_trainmodel_RNA_svmRadialWeights_scpred.rds
  threshold: 0.9

Inspection of configuration

The configuration file is in a YAML format. An example configuration can be found at the example directory.

1. dropkick

This section defines parameters for empty droplet removal by dropkick.

Parameter	Description
dropkick.skip	Skip Dropkick and use the estimated cells from Cell Ranger alone (using EmptyDrops) if set true. If set false, to estimate further empty droplets by Dropkick. Be cautious that Dropdick might predict a significant number of false negatives for a poor library.
dropkick.method	The thresholding method for labeling the training data for true cells, such as multiotsu, otsu, li, or mean.
dropkick.numthreads	Number of threads. Dropkick will use significant memory. One thread is suggested for this step.

2. filterbycount

To filter cells by nCount, nFeature, and percentage of mitochondria reads.

Parameter	Description
filterbycount.mincount	Minimum counts for a cell.
filterbycount.minfeature	Minimum features for a cell.
filterbycount.mito	Maximum percentage of mitocondria transcripts.

3. doubletfinder

This section includes three parameters for doublet removal by DoubletFinder.

Parameter	Description
doubletfinder.skip	Skip doublet detection and removal.
doubletfinder.findpK	To estimate the neighbor size (pK) by mean-variance bimodality coefficients if true. If set false, skip the estimation and use the preset pK value.
doubletfinder.numthreads	Number of threads.
doubletfinder.pK	A preset neighbor size (pK). Will be used if doubletfinder.findpK=false.

4. scpred

A pre-trained classifier for cell-type annotation by scPred.

Parameter	Description
scpred.skip	Skip the automated cell type prediction by scPred if true. This is useful for a sample without a pre-trained reference.
scpred.reference	The pre-trained reference classifier saved in a RDS file. See https://github.com/powellgenomicslab/scPred
scpred.threshold	Threshold for a positive prediction.

Result files

Three result files are generated under a result subdirectory. result/*.h5seurat and result/*.h5ad files are count matrices after processing with QC metrics such as “pANN” for proportion of artificial nearest neighbors, and/or “scpred_prediction” for predicted cell type. A report file result/report.html is a summary of QC metrics. A postproc subdirectory with postproc/*.h5ad files is also generated for basic post-processing. This includes adding a prefix to the cell barcode, ensuring unique variable names, and cleaning the raw layer from the .h5ad file.