Our goal is to identify genes that are differently expressed between clusters

• exclusively expressed in a single cluster or not

• different methods that test for:

  • differences in the mean expression level

  • differences in the rank of expression

  • differences in the proportion of cells expressing the gene

• compile a summary table

• Differential expression is comparative. Common comparisons include:

• pairwise cluster comparisons

  • eg. cluster 1 vs cluster2, cluster 1 vs cluster 3, cluster 2 vs cluster 3, etc…

• for a given cluster find ‘marker genes’ that are:

  • DE compared to at least one other cluster
  • DE compared to each of the other clusters
  • DE compared to “most” of the other clusters
  • DE and up-regulated (easier to interpret)

• cell-type comparisons (if cell type is known) - with and without clustering

findMarkers(
  sce, 
  groups = sce$louvain,       # clusters to compare
  block = sce$SampleGroup,    # covariates in statistical model
  test.type = "t",            # t-test (default)
  direction = "any",          # test for either higher or lower expression (default)
  lfc = 0,                    # null hypothesis log-fold-change = 0 (default)
  pval.type = "any"           # ranking of p-values based on any comparison (default)
)

findMarkers(
  sce, 
  groups = sce$louvain,       # clusters to compare
  block = sce$SampleGroup,    # covariates in statistical model
  test.type = "t",            # t-test (default)
  direction = "any",          # test for either higher or lower expression (default)
  lfc = 0,                    # null hypothesis log-fold-change = 0 (default)
  pval.type = "any"           # ranking of p-values based on any comparison (default)
)

• t-test: “Is the mean expression of a gene in cluster 1 and cluster 2 the same?”

• Wilcoxon rank-sum test: “It is equally likely that a randomly selected cell from cluster 1 has higher or lower expression of a gene than a randomly selected cell from cluster 2?”

• Binomial test: “Is the probability of a gene being expressed the same in cluster 1 and cluster 2?”

To an extent, all these models poorly capture the underlying features of the data.

• high noise levels (technical and biological factors)

• small library sizes

• small amounts of available mRNAs result in amplification biases and dropout events

• 3’ bias, partial coverage, uneven depth of transcripts

• stochastic nature of transcription

• multimodality in gene expression (presence of multiple possible cell states within a cell population)

However:

• t-test and Wilcoxon rank-sum test work well in practice, given at least few dozens cells to compare

• Bulk RNA-seq analysis methods do not generally perform worse than those specifically developed for scRNA-seq

• Filtering out lowly expressed genes in quite important for good performance of bulk methods (edgeR, DEseq2)

(source: Soneson & Robinson 2018)

• understand what are we trying to compare with the different tests (difference in mean expression, difference in probability of being expressed, probability of being highly/lowly expressed)

• It’s important to understand the underlying data

• It’s important to assess and validate the results

  • Strictly speaking, identifying genes differentially expressed between clusters is statistically flawed, since the clusters were themselves defined based on the gene expression data itself. Validation is crucial as a follow-up from these analyses.

• Do not use batch-integrated expression data for differential analysis

  • Instead, include batch in the statistical model (the findMarkers() function has the block argument to achieve this)

• Depending on the method you choose use: counts, normalised counts or log-normalized counts.

• Normalization strategy has a big influence on the results in differential expression.

  • e.g comparing cell types with few expressed genes vs a cell type with many genes.

• A lot of what you get might be noise. Take two random set of cells and run DE and you probably with have a few significant genes with most of the commonly used tests.

• Think of the results as hypotheses that need independent verification (e.g. microscopy, qPCR)