Dendrou group single cell pipelines

• Maintained by Charlotte Rich-Griffin
• Contributors: Charlotte Rich-Griffin, Tom Thomas and Fabiola Curion

Available pipelines:

• qc_cellranger
• qc_general_start
• integration
• clustering

Coming soon:

• demultiplexing

Introduction

These pipelines use cgat-core pipeline software

Installation:

It is advisable to run everything in a virtual environment.

e.g.

mkdir my-project
cd my-project
python3 -m venv --prompt=sc_pipelines python3-venv-scpipelines/
# This will create a my-project/venv folder

activate the environment

cd my-project
source python3-venv/bin/activate

Download and install this repo

git clone git@github.com:/DendrouLab/sc_pipelines
cd sc_pipelines
pip install --upgrade pip
pip install -r requirements.txt # installs required python packages
python setup.py develop

The pipelines are now installed as a local python package.

The pipelines use R (mostly for ggplot visualisations). The pipeline will call a local R installation (as opposed to requireing a specific build within the virtual environment) Install required R packages by copying the following code into R

install.packages(c())

Create a yml file for the cgat core pipeline software to read

vim ~/.cgat.yml

containing the following information

cluster:
    queue_manager: sge

To check the installation was successful run the following line

sc_pipelines --help

A list of available pipelines should appear!

General principles for running pipelines:

Run the pipeline for the login node on your server, it will use in built the job submission system to submit jobs.

Navigate to the directory where you want to run your analysis (this should not be within the dendrou_pipelines folder)

mkdir data_dir/
cd data_dir/
sc_pipelines qc_cellranger config

This will produce two files, pipeline.log and pipeline.yml

Edit pipeline.yml as appropriate for your data.

Then check which jobs will run with the command

sc_pipelines qc_cellranger show full

The output of this will show a list of tasks that will be run as part of the pipeline.

To run use the command

sc_pipelines qc_cellranger make full

or to run it in the background, and prevent the jobs from hanging up when you log off the server

nohup sc_pipelines qc_cellranger make full &

Occasionally you might want to run tasks individually (e.g. to debug) In order to do this you can run any task in the show full list such as:

sc_pipelines clustering make find_markers

Running the complete pipeline

Run each of pipeline qc, integration and clustering in separate folders.

1. Run sc_pipelines qc_cellranger make full
2. Use outputs to decide filtering thresholds. Note that the actual filtering occurs in the first step of integration pipeline
3. Run sc_pipelines integration make full
4. Use outputs to decide on the best batch correction method
5. Edit the pipeline yml with your preferred batch correction
6. Run sc_pipelines integration make merge_batch_correction
7. Run the clustering pipeline sc_pipelines clustering make full

Inputs to QC pipeline

There are two qc pipelines

• qc_cellranger

For qc_cellranger the minimum required columns are

sample_id	raw_path	filtered_path
Example at resources/sample_file_cellranger.txt

demultiplexing data

If you have demultiplexing data you can should include two extra columns in your samples file (e.g. resources/sample_file_inc_demultiplexing.txt); demultiplex_map_file and
demultiplex_mtd_file examples at resource/demult_map.csv and resources/demult_mtd.csv, espectively. demultiplexing_map_file should contain all barcodes, "antibody" demultimplexing_mtd_file should contain "antibody" which corresponds to the map file and any other metadata associated to your samples Other metadata that you want to add to the amnndata object can be specified in the pipeline.yml.

Note that if you are combining multiple datasets from different sources the final anndata object will only contain the intersection of the genes from all the data sets. For example if the mitochondrial genes have been excluded from one of the inputs, they will be excluded from the final data set. In this case it might be wise to run qc separately on each dataset, and them merge them together to create on h5ad file to use as input for integration pipeline.

Running pipeline modules separately

For circumstances where you arelady have a qc'd anndata object

Integration

It is possible to run the integration pipeline starting from one combined anndata object containing all your samples containing raw counts in the X slot, either with or without running filtering first. If your data is set to run simply call your anndata object [PROJECT_PREFIX]_filt.h5ad and set filtering_run: False. You must have a column called sample_id which groups the data in some way (otherwise the plotting script will break) TODO: Fix this

If you have not filtered your data then you can set run filtering_run: True, and set the remaining parameters, BUT you must ensure that your obs columns names which you want to use for filtering match the column names in resources/qc_column_names.txt

Clustering

To run clustering_scanpy without the prior steps, you will need to produce 2 anndata objects [PROJECT_PREFIX]_log1p.h5ad and [PROJECT_PREFIX]_scaled.h5ad

[PROJECT_PREFIX]_log1p.h5ad

• log normalised data in the adata.X slot
• highly variabel genes calculated

[PROJECT_PREFIX]_scaled.h5ad

• log normalised data saved in adata.raw.X
• scaled data (optionally regress) in adata.X
• pca

Minimal code:

sc.pp.normalize_total(adata, target_sum=1e4);
sc.pp.log1p(adata))
sc.pp.highly_variable_genes(adata)
adata.write("anndata_log1p.h5ad")

# optional steps
# adata = adata[:, adata.var.highly_variable]
# sc.pp.regress_out()

sc.pp.scale(adata)
sc.tl.pca(adata)
adata.write("anndata_scaled.h5ad")

If you want to use a specific batch correction then fit it into the above minimal code as appropriate

(It's probably easier to just run integration again with filtering_run: False)

Plans:

• adapt demultiplexing addition to qc to work for any per barcode metadata?

CRG 2021-02-25