Testing the interaction term in single-cell RNAseq DEG analysis

Question

[cwwong13]

I wonder if there is a way to test for the "interaction term" in Seurat. I have two categorical variables and would like to know if the interaction between these two variables is significantly different in a scRNAseq (2 x 2).

I realized that Seurat defaulted with Wilcox and the tutorial online focused on testing between two groups in one test. I know there is a method from MAST that allows more flexible modeling, which can specify random effect variables. However, the tutorials also did not showcase how to deal with interaction terms. Therefore, I wonder if there is an existing package/ method established.

I have also read the discussion here and it sounds like there is no native Seurat support for the more complex models. I am not sure if that has been fixed in the more recent updates.

Answer 1

[ATpoint]

Just get the log2 counts per million, and run limma-trend. This runs efficiently wven for large datasets, allows arbitrarily complex models and gibes full flexibility. Work from Soneson 2018 benchmarked different DE frameworks for aingle-cell data and limma performed well. See the paper and associated GitHub for recommendations on code and prefiltering of data.

https://pubmed.ncbi.nlm.nih.gov/29481549/