Gene correlation with single cell data?

Question

[theodore.killian]

I am a bioinformatician at a research institute and I frequently get requests to perform gene correlations in single cell data. For example: "we would like to correlate the expression of IP6K3 with genes involved in glutamie metabolism in endothelial cells." Insofar as I was aware, performing correlation such as Pearson or Spearman doesn't work on single data because of the high numbers of dropouts. This isn't a problem for bulk RNAseq data, where dropouts are no longer a problem, but you lose resolution in terms of cell type.

Does anyone have any recommendations for tools/packages or statistical approaches to perform gene correlation for single cell data?

[rpolicastro]

Related post

[igor]

you lose resolution in terms of cell type

With scRNA-seq, you can overcome that problem and use expression values per cell type.

[Stevens]

Fairly late answer, but I've been looking into this myself and you can observe gene-gene relationships in single cell data if you impute the missing transcripts using the R package MAGIC or another similar one. The details or on their github page - https://github.com/KrishnaswamyLab/MAGIC .

Their vignettes show you how to visualize the data but I believe you may then be able to find the correlations in the imputed count matrix using the correlatePairs function from the R package Scran

[Quang]

Stevens Would you mind elaborating on the workflow here? So you do MAGIC then scran::correlatePairs on the MAGIC output? Thank you