Skip to content

Markers identification and analyses

We have now grouped our cells according to their transcriptomic profile, we will now have to identify them biologically. If we have enough knowledge of our dataset, we can visualize the expression of the specific genes of the expected populations on the UMAP for example or in Violin Plot in order to determine which cluster will express them the most. It is also possible to identify the specific markers of each cluster by using different expression analyses.

Marker identification

Graphical identification

For example, the literature indicates that the MS4A1 gene (associated with the gene ID ENSG00000156738) is specific to B cells, so by looking at the expression of this gene on the UMAP or its distribution according to the clusters thanks to the VlnPlot function we would be able to determine which cluster would group the B cells.

To do this we will first use the FeaturePlot function which allows us to visualize our cells on a reduced dimensional space (PCA, UMAP,...) a continuous variable, it can be the expression of a gene or a continuous variable of the metadata. Then we will use VlnPlot which allows to visualize a distribution, by default it will represent a distribution by cell identity contained in active.ident (we can use the group.by parameter to use another variable).

FeaturePlot(pbmc_small,                     #SeuratObject
            features = "ENSG00000156738",   #Value to plot, can be a vector of several variable
            reduction = "umap",             #Dimensional reduction to use
            label = TRUE,                   #Plot label on the plot
            label.size = 4) +               #Change label size
  ggtitle(annotated_hg19[annotated_hg19$ensembl_gene_id == "ENSG00000156738", "external_gene_name"],
          "ENSG00000156738")


VlnPlot(pbmc_small,                         #SeuratObject
        features = "ENSG00000156738") +     #Variable to plot
  ggtitle(annotated_hg19[annotated_hg19$ensembl_gene_id == "ENSG00000156738", "external_gene_name"],
          "ENSG00000156738")

With these results we can consider cluster 3 as being composed of B cells.

Differential expression analysis

It is however sometimes difficult to use this method for each of our clusters. Seurat proposes to identify the specific markers of each cluster using a differential expression analysis method.

For each cluster and each gene, the FindAllMarkers function will determine if there is a significant difference between the gene expression of the cells in our cluster and the other cells. By default, it uses the non-parametric Wilcoxon Rank Sum test (also called Mann-Whitney). He then performs a Bonferroni multiple correction test.

Here we have changed some parameters to not filter any gene which will be very useful for the enrichment analysis (GSEA) which is based on an ordered list of genes.

pbmc_markers <- FindAllMarkers(pbmc_small,              #SeuratObject
                               only.pos = FALSE,        #Returns positive and negative gene markers
                               min.pct = 0.1,           #Take into account genes that are detected in at least 10% of the cells
                               logfc.threshold = 0,     #Return markers with a logFC superior to threshold
                               test.use = "wilcox",     #Method used
                               verbose = FALSE)

## Preview of the resulting dataframe
kable(head(pbmc_markers))
p_val avg_log2FC pct.1 pct.2 p_val_adj cluster gene
ENSG00000137154 0 0.6952948 0.998 0.997 0 0 ENSG00000137154
ENSG00000144713 0 0.6444963 0.998 0.997 0 0 ENSG00000144713
ENSG00000112306 0 0.7383383 1.000 0.994 0 0 ENSG00000112306
ENSG00000177954 0 0.7437027 0.998 0.994 0 0 ENSG00000177954
ENSG00000164587 0 0.6352699 1.000 0.997 0 0 ENSG00000164587
ENSG00000118181 0 0.7973422 1.000 0.978 0 0 ENSG00000118181

The result of this function is a dataframe with several columns:

  • p_val : p-value of the statistical test used
  • avg_log2FC : log2(Fold change +1) between the average expression of the considered cluster and the average expression of the rest of the cells
  • pct.1 : percentage of detection of the gene in our cluster
  • pct.2: percentage of detection of the gene in the rest of the cells
  • p_val_adj : adjusted p-value (Bonferroni correction)
  • cluster : cluster considered
  • gene : name of the gene

Warning

Be careful not to take into account the names of the lines in this dataframe for reference. Indeed, it is quite frequent that a gene is defined as a marker for several clusters which will duplicate the line names and thus add suffixes in the rows. We would be back to the same problem as if we were using gene names in Read10X.