marionilab / deconvolution2016

Code and manuscript files for Aaron and Karsten's deconvolution paper.

Using deconvolution to normalize scRNA-seq data with many zeroes

Overview

This repository contains the code and manuscript files for the paper Pooling across cells to normalize single-cell RNA sequencing data with many zero counts, by Lun et al. (2016).

Note: Further updates and development of the analysis and simulation code will take place at https://github.com/MarioniLab/FurtherNorm2018. If you have general questions regarding the code (i.e., not specifically involving the manuscript), please post your issues at the above repository instead.

Simulations

To run the simulation code, enter simulations and then:

1. Run lowcounts.R to perform the low-count simulations, or brittlesim.R to perform the high-count simulations.
2. Run standerr.R to estimate the variance of the size factor estimates across methods.
3. Run poolsim.R to compare the variability of the estimates with and without the ring arrangement.
4. Run complexity.R to determine the time-complexity of the deconvolution method.

You can also run fewcounts.R to see behaviour with few cells, or highcounts.R to see behaviour at very high counts.

Real analyses

To run the real data analysis code:

1. Make a data subdirectory and download the Zeisel et al. tables (http://linnarssonlab.org/cortex) and the Klein data (supplementary tables in GSM1599494, GSM1599499).
2. Enter the realdata directory and run Zeisel.R and Klein.R to pre-process the data and estimate size factors for all cells in each of those two data sets.
3. Run edgeR.R to identify DE genes in each data set, and GOAnalysis.R to perform a GO analysis on the DE genes.
4. Run HVGAnalysis.R to identify highly variable genes in each data set.
5. Run switchTestedgeR.R to perform the offset/covariate switching analysis.

Also, run plotKleinParam.R to generate plots that justify parameter settings in the simulations.

Manuscript

The manuscript directory contains all LaTeX code used to generate the manuscript. This can be compiled with make. It assumes that all of the simulations and real data analyses have already been performed.

deconvolution2016's Issues

What to put in Supps as discussed on 9/12/15

The following sections will be moved into the supplements:

• In depth explanation of the deconvolution method 3.3-3.5 (p6-8)
• Spike-In stuff 4.4 (p12-13)

Shuffle miscellaneous text

Need to move timings into "constructing the linear system" section. Or maybe I should put it into the "obtaining sensible least squares" section, because the cubic complexity comes from solving the system. Something like:

The computational complexity of devolution is determined by the number of cells involved in the construction and solution of the linear system. This is discussed in more detail in Section X of Additional File 1. Empirical timings are also provided in Supplementary Figure Y.