Single-cell RNA sequencing
Single cells from mouse cortex, analysed from quality control to cell types, with conditions compared using animals as replicates.
Guided experiment: Finding cell types with single-cell RNA sequencing.
Single-cell RNA sequencing
Single-cell RNA sequencing counts the messenger RNA molecules in thousands of individual cells, so that cell types can be recognised from the genes they express. Cells are dissociated from tissue and captured one by one; their transcripts are tagged with a cell barcode, amplified and sequenced. Applied to the mouse cortex, the method has classified cells into many molecularly distinct types with characteristic marker genes (Zeisel et al., 2015; Tasic et al., 2016). Analysis follows an established sequence of quality control, normalisation, feature selection, dimensionality reduction, clustering, annotation and differential testing (Luecken and Theis, 2019).
Design the experiment
Questions to investigate
- Set the quality-control thresholds. Which damaged cells, doublets and empty barcodes does each threshold remove, and how many intact cells are lost with them?
- Change the number of clusters. When are cell types merged, and when are they split?
- Annotate each cluster from its marker genes. How accurate is the annotation, and which clusters are hardest to label?
- Compare cell-type proportions between conditions with animals as replicates, and then with cells as replicates. Why are the cell-level p-values so much smaller, and why are they invalid?
- Process all controls on one day and all treated animals on another, with a strong batch effect and no treatment effect. What does the comparison then show?
- How many animals per condition are needed to detect the change in proportion, and the change in gene expression within that cell type?
In a real laboratory
Mice are allocated to conditions at random, and the number of animals is planned from the variation between animals, because the animal, not the cell, is the experimental unit. The cortex is dissected quickly into cold buffer and dissociated. Viability and cell concentration are checked before droplet capture, because damaged cells lower data quality and loading more cells raises the doublet rate. Libraries carry cell barcodes and unique molecular identifiers and are sequenced to a planned depth. Samples from every condition are processed on each day, in random order, so that batch is not confounded with condition. Quality-control thresholds and the analysis plan are fixed in advance, and raw reads and count matrices are deposited in a public archive. All procedures require approval by an Institutional Animal Ethics Committee (in India, under the CCSEA and the Prevention of Cruelty to Animals Act 1960; elsewhere, the equivalent national body) and follow the 3Rs.
How the simulation works
Twelve cortical cell types are simulated with standard marker genes, together with 13 mitochondrial genes and neutral genes (1,200 genes in total, so counts per cell are lower than in a whole-transcriptome experiment). Counts are drawn from a negative binomial distribution around a cell-specific library size that rises with sequencing depth and saturates. Damaged cells have smaller libraries and about 15–60% mitochondrial reads, doublets are the sum of two cells, and some barcodes contain only ambient RNA. Animals differ in cell-type proportions; the treatment changes the proportion of one type and six of its genes; processing days differ in neuron recovery, capture efficiency, damaged cells and 40 genes. The analysis normalises to 10,000 counts per cell and log(1 + x), selects highly variable genes, computes principal components, clusters by k-means, finds markers by Wilcoxon rank-sum test with Benjamini–Hochberg correction, and compares proportions per animal by Welch t-test. The two-dimensional plot shows principal components, not UMAP or t-SNE. Ambient contamination of intact cells, other differences in expression between animals, cell-cycle effects, graph-based clustering and dedicated doublet removal are not simulated, and pseudobulk testing uses a Welch t-test rather than count models, so its power is approximate.
- Zeisel, A., Muñoz-Manchado, A. B., Codeluppi, S., et al. (2015). Cell types in the mouse cortex and hippocampus revealed by single-cell RNA-seq. Science, 347(6226), 1138–1142. https://doi.org/10.1126/science.aaa1934
- Tasic, B., Menon, V., Nguyen, T. N., et al. (2016). Adult mouse cortical cell taxonomy revealed by single cell transcriptomics. Nature Neuroscience, 19(2), 335–346. https://doi.org/10.1038/nn.4216
- Luecken, M. D., & Theis, F. J. (2019). Current best practices in single-cell RNA-seq analysis: a tutorial. Molecular Systems Biology, 15(6), e8746. https://doi.org/10.15252/msb.20188746
About the simulations
Each simulator is a simplified version of a published model of cells, signals, instruments or behaviour, and runs in the web browser. The simulations show how the techniques work and how design choices change the results; they do not replace training in a laboratory.
All data are simulated and generated afresh on each run, so results vary between runs, as in real experiments. Nothing is stored or sent: designs and results are lost when the page is closed, unless the protocol is copied or the data file is downloaded.
Last updated