Assessing the reliability of spike-ins for scRNA-seq data

Overview

This repository provides code for the paper Assessing the reliability of spike-in normalization for analyses of single-cell RNA sequencing data by Lun et al. (2017).

Repeating the real data analysis

Run download.sh to download the count matrices from ArrayExpress (E-MTAB-5522), unpack them and move them to the relevant directories. This will also download the SDRF file describing the annotation for all samples.

Install package/ using R CMD INSTALL --preclean package (or an equivalent command for your installation). This was last tested with R version 3.4.* and Bioconductor version 3.6.

Enter real and follow these instructions:

1. Enter Calero/trial_20160113/ and run run_me.sh. This will produce a Markdown document containing the variance estimates, along with some serialized R objects for further inspection if necessary.
2. Repeat for Calero/trial_20160325, Liora/test_20160906 and Liora/test_20170201.
3. Run make_pics.R to reproduce the figures in the paper.
4. Enter depth/ and run runner.R to perform the simulations for sampling noise.
5. Enter index_swapping/ and run check_swap.R to generate the figures checking for index swapping.

Repeating the simulations

To run the simulations, enter the simulations directory:

1. Enter datasets/ and run download.sh to download the various data files. Also run prerunner.R to pre-process them into serialized R objects.
2. Enter sampling/ and run resampler.R to simulate sampling noise. You can also run sizevar.R to quantify the variation in size factors across cells.
3. Enter variance/ and run varsim.R to perform the simulations for detecting HVGs.
4. Enter diffexp/ and run desim.R to perform the simulations for detecting DEGs.
5. Enter clustering/ and run clustsim.R to perform the simulations for clustering and PCA. This requires you to obtain pancreas_refseq_rpkms_counts_3514sc.txt.gz from the processed files in E-MAT-5061, along with the corresponding SDRF file.
6. Enter pics/ and run picmaker.R and get_properties.R to reproduce the figures in the paper.

The sequences/biophysical/ directory contains a script to examine the differences in biophysical properties between the spike-in and endogenous mouse genes.

Realigning and recounting

Readers interested in regenerating the count matrices from the FASTQ files are advised to:

1. Follow the instructions in sequences/genomes/README.md to build the genome indices. Similarly, follow the instructions in sequences/annotation/README.md to obtain the annotation.
2. Download the FASTQ files from ArrayExpress. Files corresponding to each batch of data should be placed in Calero/trial_20160113/fastq, etc.
3. Run the various mapme.sh scripts to execute the master scripts for alignment, and count_me.sh for read counting. Paths should refer to the top-level tools/ directory obtained using download.sh.

Note that the mapme.sh scripts in Liora need to be modified to retrieve file names from the fastq subdirectory.

Creating the manuscript

The manuscript directory contains all LaTeX code used to generate the manuscript. This can be compiled with make.