Which values from Seurat::FetchData function are to be used for correlation analysis between genes?

Question

I want to perform a correlation test between genes in on my single cell RNA seq data set. I perfomed the differential expression analysis using the Seurat version 2 package, after performing stages of normalisation, scaling, PCA, TSNE analyses and clustering. My correlation plot is far from what I expect and I want to know if the expression values obtained with Seurats V2's FetchData function (imputed, scaled, or raw) may have an effect on correlation analysis' outcome. This is what I did:

# load Seurat v2
library(Seurat)

# load corrplot package
library(corrplot)

# obtain expression values of the genes I want to test rom the Seurat object
expression_matrix <- FetchData(object, vars.all = c("gene1", "gene2", "gene3"))

expression_matrix <- round(expression_matrix , 1)

cor_expression_matrix <- cor(expression_matrix)

corrplot(cor_expression_matrix , is.corr = T, method = "square")

Which gene expression values should be best used for the analysis of correlation between genes: imputed, scaled, or raw?

Thank you in advance for your kind response.

Answer 1

You should use normalized data. The function has a parameter slot="data" that let's you select to access the normalized data.

Any reason for rounding to 1 decimal place here? I imagine that would have a substantial impact on the correlation you compute.