single cell mixology: single cell RNA-seq benchmarking

note: the repository has been renamed to sc_mixology. the old link will be redirected to the current repository.

sc_mixology uses three human lung adenocarcinoma cell lines HCC827, H1975 and H2228, which were cultured separately, and then processed in three different ways. Firstly, single cells from each cell line were mixed in equal proportions, with libraries generated using three different protocols: CEL-seq2, Drop-seq (with Dolomite equipment) and 10X Chromium. Secondly, the single cells were sorted from the three cell lines into 384-well plates, with an equal number of cells per well in different combinations (generally 9-cells, but with some 90-cell population controls). Thirdly, RNA was extracted in bulk for each cell line and the RNA was mixed in 7 different proportions and diluted to single cell equivalent amounts ranging from 3.75pg to 30pg and processed using CEL-seq 2 and SORT-seq. ERCC spike-in controls were present in samples processed using the 2 plate-based technologies (CEL-seq2 and SORT-seq).

Raw data from this series of experiments is available under GEO accession number GSE118767. The processed count data obtained from scPipe is stored in R objects that use the SingleCellExperiment class. Below are instructions for getting the count data and metadata (including annotations) for each dataset. All data is post sample quality control, without gene filtering.

Summary of all datasets

Load files into R

You can find R object files in the data folder

load("data/sincell_with_class.RData")

This will create three variables: sce10x_qc, sce4_qc, and scedrop_qc_qc. sce10x_qc contains the read counts after quality control processing from the 10x platform. sce4_qc contains the read counts after quality control processing from the CEL-seq2 platform. scedrop_qc_qc contains the read counts after quality control proessing from the Drop-seq platform.

ground truth

The true label is stored in colData(). For single cells the label is in column cell_line_demuxlet. For single cell mixtures the ground truth is the combination of three cell lines, which is in column H1975, H2228 and HCC827. so one merge and use the combination as the label, such as paste(sce_SC1_qc$H1975,sce_SC1_qc$H2228,sce_SC1_qc$HCC827,sep="_"). Similarly, the ground truth in RNA mixture is the proportion of RNA from each cell line, stored in column H2228_prop, H1975_prop and HCC827_prop, which can be merged into one column and use as the label, such as paste(sce2_qc$H2228_prop,sce2_qc$H1975_prop,sce2_qc$HCC827_prop,sep="_").

Counts

To access count data from a SingleCellExperiment object, use the counts(sce) function:

counts(sce10x_qc)[1:5, 1:5]

Metadata

To access sample information from a SingleCellExperiment object, use the colData(sce) function:

head(colData(sce10x_qc))

Examples of using these datasets

You can find an Rnotebook in the script/data_QC_visualization folder named data_explore_mixture.Rmd which includes code for analysing the cell mixture and RNA mixture datasets.

Scripts for reproducing a broader methods comparison

The [script] folder contains scripts that can reproduce the analysis and figures from our paper: Benchmarking single cell RNA-sequencing analysis pipelines using mixture control experiments.

Note: The ggtern package, which has been used to generate the ternary plots, has known issues with recent versions of ggplot and the relevant code may be broken if you have updated the ggplot package.