Differential Expression Analysis with monocle and batch effect correction

Question

[Poorya Parvizi]

Hi,

I would like to run differential expression analysis on my single-cell data. My data contains of 5 groups which sequenced in 3 batches. Each batch contains all of the groups. Before the differential expression analysis with monocle (I should run it with monocle) I have tried to eliminate batch effect from the data. I used Seurat's ScaleData function (vars.to.regress argument) to regress out the batch effect. However, the output of this function is in scale format which contains negative values in it.

Can I use scale values in differential expression analysis, specifically monocle?

I also tried limma's removeBatchEffect function, which also gives negative values. What is the best way to regress out the batch effect before the monocle differential expression?

Answer 1

[Kevin Blighe]

I do not believe negative values will help differential expression analysis in any way. Monocle actually provides functionality for dealing with things like batch. In many of the functions, there is a parameter called residualModelFormulaStr, which allow you to list any covariates for which the statistical modelling should be adjusted.

residualModelFormulaStr

A model formula string specify effects you want to exclude when testing for cell type dependent expression

So, for example, for a differential expression analysis, use:

differentialGeneTest(cds, fullModelFormulaStr = " ~ condition + batch",
reducedModelFormulaStr = " ~ batch", relative_expr=TRUE, cores=4)

Take a look at the Manual and Tutorial.

Kevin

[Poorya Parvizi]

Thank you Kevin. I will try it.

[zhangguoqiang86]

I believe it should be

differentialGeneTest(cds, fullModelFormulaStr = " ~ condition + Batch", reducedModelFormulaStr = " ~ Batch", relative_expr=TRUE, cores=4)

according to here: Multi-Factorial Differential Expression Analysis

[Kevin Blighe]

Thank you - I have edited the answer.