Skip to main content

16  Group projects

NoteNotes
  • 14:00 - 18:00: project work, in groups
  • 18:00 - 20:00: group presentations (10’ each group, max 5 slides: what / why / where / when / how, conclusions)
TipAims
  • Analyse a single-cell RNA-seq dataset by yourself, from A to Z
  • Make (and justify!) your own analysis choices
  • Present and discuss your results with the other participants

16.1 The dataset: multiciliated cell differentiation

Multiciliated cells (MCCs) line the brain ventricles, the airways and the reproductive tracts: each of them grows dozens to hundreds of motile cilia, which beat to move fluids. MCCs differentiate from progenitors which exit the cell cycle, become post-mitotic, then amplify their centrioles (in deuterosomal cells) to template the motile cilia of mature MCCs.

Each group will analyse scRNA-seq datasets of the in vitro differentiation of mouse radial glial cell progenitors into multiciliated cells:

  • WT: 2 replicates (4,111 + 11,316 filtered cells), from two different genetic backgrounds (C57Bl6J and C57Bl6N);
  • Mutant: a single replicate (6,639 filtered cells, C57Bl6J background).

The data is available in your session, in the ~/Share/group_exercises/ folder: for each dataset, a count matrix (*_counts.mtx.gz, in Matrix Market format), the genes (*_genes.tsv.gz) and the cells (*_barcodes.tsv.gz, with their sample, genetic background, total UMI counts and number of detected genes).

bash
ls -lh ~/Share/group_exercises/

Import both datasets in R as SingleCellExperiment objects (this takes ~30 seconds and ~4 GB of memory):

R
library(SingleCellExperiment)
read_sample <- function(sample, dir = "~/Share/group_exercises") {
    counts <- Matrix::readMM(file.path(dir, paste0(sample, "_counts.mtx.gz")))
    genes <- readr::read_tsv(file.path(dir, paste0(sample, "_genes.tsv.gz")))
    cells <- readr::read_tsv(file.path(dir, paste0(sample, "_barcodes.tsv.gz")))
    sce <- SingleCellExperiment(
        assays = list(counts = as(counts, "CsparseMatrix")),
        rowData = genes,
        colData = cells
    )
    rownames(sce) <- rowData(sce)$ID
    ## The two WT replicates share some barcodes: prefix them with the
    ## sample and genetic background
    colnames(sce) <- paste(sce$sample, sce$background, sce$barcode, sep = "_")
    sce
}
WT <- read_sample("WT")
mutant <- read_sample("mutant")
WT
mutant
table(WT$background)

16.2 Suggestions

  • Is there a strong batch bias in the WT dataset?
  • Can a trajectory be inferred from the in vitro WT differentiation scRNA-seq experiment?
  • Annotate the putative cell cycle phase of each cell in the WT dataset, using the algorithm of your choice.
  • Which cyclins are expressed in differentiating cells?
  • Which cyclins are missing in differentiating cells in the mutant dataset, compared to the WT dataset?

16.3 Analyse it

Here is a checklist of the analysis steps we covered during the week. Not all of them will be relevant to your questions: decide which ones are, and why.

  1. Import the data into a SingleCellExperiment object (Lab 3, Lab 4)
  2. Quality control and filtering of cells and genes (Lab 4)
  3. Normalization, feature selection and dimensionality reduction (Lab 5)
  4. Batch correction, if required (Lab 6)
  5. Clustering, marker genes and cell annotation (Lab 5)
  6. Differential expression analysis between conditions, with pseudobulk (Lab 6)
  7. Trajectory inference and pseudotime (Lab 8)

16.4 Present it

Prepare a short presentation (10 minutes, max 5 slides), covering:

  • What: what is the biological question, and what is the dataset?
  • Why: why is single-cell RNA-seq relevant to answer this question?
  • Where / When / How: which analysis steps did you perform, and which choices did you make?
  • Conclusions: what did you learn, and what would you do next?

16.5 Working on another dataset

Groups who would rather analyse another dataset (e.g. from their own research, or from a paper they like) can do so: discuss it with the instructors first. The scRNAseq package distributes many processed datasets:

R
library(scRNAseq)
datasets <- surveyDatasets()
datasets
searchDatasets("pancreas")
sce <- fetchDataset("baron-pancreas-2016", "2023-12-14", path = "human")

The CZ CELLxGENE Discover and Human Cell Atlas portals, as well as GEO, are other good places to find processed count matrices.