3. Feature selection

Because of the sparsity in the sequencing data many genes or features are almost no expressed. Additionally, some genes are constantly expressed across cells. These features are then probably not playing any function in cells and on the other hand can just add noise and unnecessary complexity to further analysis. Then, it's usual to remove genes with very low variability and to select only top highly variable genes (HVG).

We will use the function FindVariableFeatures() to calculate the top most variable genes. The parameter nfeatures is used to set the number of top selected genes. We set to the top 1000 features.

pbmc.filtered <- FindVariableFeatures(pbmc.filtered, nfeatures = 1000)

We can access to the top 1000 variable features using the VariableFeatures function. In the next chunk we display the top first 6 (head) of this set.

head(VariableFeatures(pbmc.filtered))

## [1] "S100A9" "LYZ"    "S100A8" "FTL"    "GNLY"   "FTH1"

In the next scatter plot we can see the average expression vs the standardized variance for each feature. Genes in red are the selected HVG.

# plot variable features with and without labels
plot1 <- VariableFeaturePlot(pbmc.filtered)
plot1 <- LabelPoints(plot = plot1, 
                     points = head(VariableFeatures(pbmc.filtered),
                                   10), 
                     repel = TRUE)
plot1 

For further analysis we will use only HVGs.

Exercises

Extract normalized expression values and calculate the variance in genes and order them by variance.

• Extract the matrix of gene expression normalized values from the Seurat object
• Calculate variances manually from the matrix
• Sort genes based on variances in decreasing order and show top 6 genes

Previous Chapter| Next Chapter