Question

EnhancedVolcano and scRNAseq differential gene expression

1

Entering edit mode

2.7 years ago

TJ ▴ 50

Hi,

I'm looking to make a volcano plot of differentially expressed genes between two groups of cells from a scRNAseq dataset that was analyzed using Seurat.

Here's code used to generate the DE genes. The seurat object combined is an integrated object with ActiveAssay(combined) = "RNA".

my.deg <- FindMarkers(combined, 
                            ident.1 = c("1", "2"), 
                            ident.2 = c("0", "3", "4"), 
                            verbose = FALSE)

This creates a data.frame with gene names as rows, and includes avg_log2FC, and adjusted p-values. This is done using the Seurat FindMarkers function default parameters, which to my understanding uses a wilcox.test with a Bonferroni correction.

Next, I'm looking to visualize this using a volcano plot using the EnhancedVolcano package:

EnhancedVolcano(my.deg , 
                rownames(my.deg ),
                x ="avg_log2FC", 
                y ="p_val_adj")

I get a plot, but I see a straight line across the top on one side with a bunch of genes on top of each other. I suspect this is from the input data.frame as there are a bunch of 0s in the p_val_adj column. Is there a way to deal with this? I know EnhancedVolcano sets there to 10^-1, but this still shows a bunch of genes on top of each other on a line. Any guidance on how to fix/adjust this would be much appreciated.

enter image description here

seurat wilcox expression gene test EnhancedVolcano differential scRNAseq • 10k views

ADD COMMENT • link 2.7 years ago by TJ ▴ 50

score 4 · Accepted Answer · 2022-03-10

4

Entering edit mode

2.7 years ago

andrew.j.skelton73 6.6k

This is part of the statistics... These genes being differentially tested with a Wilcoxon test show an FDR value of 0, which on the -log10 scale is Inf. As you suggest this is likely fixed as a value in EnhancedVolcano, hence why you see the plateau at the top. Ultimately, there is nothing to fix really.

On a more right thing to do note, I'd suggest you explore pseudo bulk-ing this data around your biological replicates as your chance of false positive could be quite high.

ADD COMMENT • link 2.7 years ago by andrew.j.skelton73 6.6k

2

Entering edit mode

It's probably the machine limit towards small values -log10(.Machine$double.xmin) which would be 307.6527 so that fits the plot.