IRescue 1.2.0

Uncertainty-aware quantification of transposable elements expression in scRNA-seq

IRescue - Interspersed Repeats single-cell quantifier

IRescue quantifies the expression fo transposable elements (TEs) subfamilies in single cell RNA sequencing (scRNA-seq) data, performing UMI-deduplication with sequencing errors correction (for 10X-like libraries) or read quantification (for UMI-less libraries, e.g. SMART-seq) followed by probabilistic assignment of multi-mapping reads by an Expectation-Maximization (EM) procedure. The output is written on a sparse matrix compatible with Seurat, Scanpy and other toolkits.

Content

• Installation
  • Using conda
  • Using pip
  • Install a pre-release version
  • Build from source
  • Container (Docker/Singularity)
• Usage
  • Required inputs
  • Custom annotation
  • UMI-less libraries (e.g. SMART-seq)
  • Best practices
  • Output files
  • Load IRescue data with Seurat
• Cite

Installation

Using conda (recommended)

Use conda (or mamba or micromamba) to install IRescue with all its dependencies.

conda create -n irescue -c conda-forge -c bioconda irescue

Using pip

If installing with pip, the following requirements must be installed manually: python>=3.9, samtools>=1.12, bedtools>=2.30.0, and fairly recent versions of the GNU utilities are required (tested on gawk>=5.0.1, coreutils>=8.30 and gzip>=1.10).

pip install irescue

Install a pre-release version

You can install or upgrade to a pre-release using pip:

pip install -U --pre irescue

Build from source

By building the package directly from the source, you can try out the features and bug fixes that will be implemented in the future release. As above, you need to install some requirements manually. Be aware that builds from the development branches may be unstable.

git clone https://github.com/bodegalab/irescue
cd irescue
pip install .

Container (Docker/Singularity)

Docker and Singularity containers are available for each conda release of IRescue. Choose the TAG corresponding to the desired IRescue version from the Biocontainers repository and pull or execute the container with Docker or Singularity:

# Get latest biocontainers tag (with curl and python3, otherwise check the above link for the desired version/tag)
TAG=$(curl -s -X GET https://quay.io/api/v1/repository/biocontainers/irescue/tag/ | python3 -c 'import json,sys;obj=json.load(sys.stdin);print(obj["tags"][0]["name"])')

# Run with Docker
docker run quay.io/biocontainers/irescue:$TAG irescue --help

# Run with Singularity
singularity exec https://depot.galaxyproject.org/singularity/irescue:$TAG irescue --help

Usage

Inspect all parameters:

irescue --help

Quick start:

irescue -b genome_alignments.bam -g hg38

Required inputs

BAM file sorted by coordinate, indexed and annotated with cell barcode and, optionally, UMI sequences as tags (e.g. CB and UR tags, configurable with --cb-tag and --umi-tag)

It can be obtained by aligning reads using STARsolo. It is highly recommended to keep secondary alignments in BAM file, that will be used in the EM procedure to redistribute multi-mapping reads (at least --outFilterMultimapNmax 100 --winAnchorMultimapNmax 100), and remember to output all the needed SAM attributes (e.g. --outSAMattributes NH HI AS nM NM MD jM jI XS MC ch cN CR CY UR UY GX GN CB UB sM sS sQ).

Custom annotation

A custom repeats annotation can be provided in BED format (e.g. -r TE.bed) of at least four columns, with the fourth column being the TE feature name (e.g. subfamily name).

UMI-less libraries (e.g. SMART-seq)

Only in pre-release version 1.2.0b2 or later.

You can ignore the UMI sequence (thus skipping UMI-deduplication entirely) with --no-umi:

irescue -b genome_alignments.bam -g hg38 --no-umi

NB: the BAM tag for cell barcodes sometimes is RG instead of CB. In such case, add the parameter --cb-tag RG.

Best practices

• If you already obtained gene-level counts (using STARsolo, Cell Ranger, Alevin, Kallisto or other tools), it is advised to provide the whitelisted cell barcodes list as a text file (-w barcodes.tsv). This will significantly improve performance by processing viable cells only.

• For optimal run time, use at least 4-8 cpus, e.g.: -p 8.

Output files

IRescue generates TE counts in a sparse matrix readable by Seurat or Scanpy into a counts/ subdirectory. Optional outputs include a description of equivalence classes with UMI deduplication stats ec_dump.tsv.gz and a subdirectory of temporary files tmp/ for debugging purpose (only kept with the --keeptmp parameter). You can enable a highly detailed logging with -vv (printed to the terminal's stderr).

irescue_out/
├── counts/
│   ├── barcodes.tsv.gz
│   ├── features.tsv.gz
│   └── matrix.mtx.gz
├── ec_dump.tsv.gz
└── tmp/

Load IRescue data with Seurat

Multiple assays can be exploited to integrate TE counts into an existing Seurat object containing gene expression data:

# import TE counts from IRescue output directory
te.data <- Seurat::Read10X('./IRescue_out/', gene.column = 1, cell.column = 1)

# create Seurat assay from TE counts
te.assay <- Seurat::CreateAssayObject(te.data)

# subset the assay by the cells already present in the Seurat object (in case it has been filtered)
te.assay <- subset(te.assay, colnames(te.assay)[which(colnames(te.assay) %in% colnames(seurat_object))])

# add the assay in the Seurat object
seurat_object[['TE']] <- irescue.assay

The result will be something like this:

An object of class Seurat 
32276 features across 42513 samples within 2 assays 
Active assay: RNA (31078 features, 0 variable features)
 1 other assay present: TE

From here, TE expression can be normalized. To normalize according to gene counts or TE+gene counts, normalize manually or merge the assays. Reductions can be made using TE, gene or TE+gene expression.

Cite

Benedetto Polimeni, Federica Marasca, Valeria Ranzani, Beatrice Bodega, IRescue: uncertainty-aware quantification of transposable elements expression at single cell level, Nucleic Acids Research, 2024; https://doi.org/10.1093/nar/gkae793