CD8_clustering

Background and Rationale

Single cell RNA-seq (scRNA-seq) experiments are increasing in prevalence within biological experimentation. An advantage of scRNA-seq over bulk sequencing experiments is the ability to look at gene expression of individual cells. This level of granularity allows for analysis that would otherwise not be possible with bulk sequencing experiments.

A classic example of single cell specific analysis is cell clustering. In a bulk sequencing experiment, the gene expression data is a mixture of various cell types that cannot be separated, so it is extremely difficult to get an accurate picture of the cellular composition of a sample. In scRNA-seq experiments, individual cells within a sample can be separated into specific sub-type clusters based on their gene expression profiles using classic clustering methods, resulting in the ability to infer the composition of cellular sub-types of a sample.

In this CD8_clustering workflow, we look specifically at CD8+ T cells, also known as "killer T cells", which have a cytotoxic function within adaptive immunity. CD8+ T cells are typically categorized into specific subtypes - naive, memory (stem cell, central, and effector), effector, and exhausted. The subtypes of CD8+ T cell have different gene expression profiles as well as different functions in the immune system.

By performing clustering on scRNA-seq data of CD8+ T cells, one can infer the subtypes of individual cells based on the characteristic gene expressions of each cluster and gain valuable information on the cellular composition of given sample(s) and therefore the immune function of cells.

Clustering Method

This workflow makes use of the K Nearest Neighbours clustering algorithm (KNN) where the number of nearest neighbours to use can be user specified. Since scRNA-seq data is highly dimensional and sparse, dimensionality reduction is necessary in order to performing clustering. The most common dimensionality reduction technique is Principal Component Analysis, but other methods are also available in this workflow (t-SNE and UMAP).

Workflow

The main steps of this workflow are:

1. FTP download of raw gene expression matrix files for all relevant samples from the NCBI GEO (tar format).
2. Extract individual sample specific data files (in this DAG, there are 7 individual samples).
3. Aggregation of individual sample data into one singular gene expression matrix.
4. Filter features and cells based on user determined thresholds.
5. Clustering of cells using KNN on dimension reduced gene expression data.

Usage

First, clone the git repository to your local machine and enter the project directory:

git clone git@github.com:pattiey/CD8_clustering.git
cd CD8_clustering

Ensure that necessary channels are available:

conda config --add channels 'bioconda'
conda config --add channels 'r'
conda config --add channels 'conda-forge'

Then create an environment with the required packages:

conda create --name cd8_clustering --file env.txt

Activate the environment:

conda activate cd8_clustering

Due to package conflicts, local R will be used and R packages must be installed outside of the Conda environment. Please ensure that your R version is up to date. To install needed R packages, run:

Rscript /path/to/CD8_clustering/scripts/init.R

And enter yes for any prompts that may appear.

Run the Snakemake workflow with relevant parameters. Be sure to set the appropriate number of cores. Ensure that the config.yaml file is updated with the relevant fields before running.

snakemake --snakefile /path/to/CD8_clustering/Snakemake/Snakefile --configfile /path/to/CD8_clustering/Snakemake/config.yaml --cores 4

Input

The input for this workflow is controlled by the config.yaml file. Here is a description of the fields of the config.yaml file.

Field	Description
FTP_URL	FTP download file from NCBI GEO of raw gene expression data
DATA_DIR	Directory where data is to be stored
SCRIPTS_DIR	Directory where project scripts are stored
OUTPUT_DIR	Directory where output files are to be stored
PROJECT	Name of experiment/project
SAMPLES	Sample names from GEO

Other user specified parameters are also available to adjust in the Snakefile.

Snakemake rule	Parameter	Description
filter_cells	mito	maximum percentage threshold of mitochondrial expression to filter
filter_cells	ribo	maximum percentage threshold of ribosomal expression to filter
filter_cells	nFeature_lo	minimum number of features present in cells to keep
filter_cells	nFeature_hi	maximum number of features present in cells to keep
filter_cells	nCount_lo	minimum number of counts required to keep a cell
filter_cells	nCount_hi	maximum number of counts required to keep a cell
cluster_cells	reduction	method of dimensionality reduction for clustering, pca, tsne, or umap
cluster_cells	k	Number of neighbours to use for K nearest neighbours clustering
cluster_cells	num_features	Number of features to use for SCTransform
cluster_cells	resolution	Value of the resolution parameter, use a value above (below) 1.0 if you want to obtain a larger (smaller) number of communities.
cluster_cells	dims	Number of dimensions to use for clustering.

The example config file in this repository contains the samples and URL data from the NCBI Gene Expression Omnibus (GEO) GSE116390. This is single cell RNA-seq data of CD8+ T cells from B16 melanoma tumours from mice pertaining to the experiments done by S. Carmona et al..

Output

The workflow produces the result of KNN clustering on the scRNA-seq samples.

Output File	Description
cellClusters.csv	A CSV file containing cluster labels of each cell identified through barcode and sample
PCA_plot.png	A plot of the first two principal components of the gene expression data, coloured by cluster
TSNE_plot.png	A t-SNE plot of the gene expression data, coloured by cluster
UMAP_plot.png	A UMAP plot of the gene expression data, coloured by cluster
cell_comp.png	A plot of cluster proportions, total and by sample

Using the sample data and the parameters specified in the Snakefile, here are the plots produced by the workflow.

The K Nearest Neighbours clustering found four distinct clusters. Using this clustering, one could then perform further analysis to find the corresponding CD8+ cell subtype for each cluster in order to extract information about the cellular composition and immuno-landscape of the experimental samples.

Principal Component Plot

t-SNE Plot

UMAP plot

Cluster Composition Plot