Question about logFC while using DESeq2 in Seurat

Question

[suvratha]

Hello, the question I have is - log to which base is reported when we use test.use="DESeq2" in the FindMarkers() function?

I know the default logFC while using Wilcoxon test in FindMarkers() is reported in the natural log, is this true when we use DESeq2 as well? I have this doubt as DESeq2 reports FC in log to the base 2 and not natural log.

Thanks Suvi

Answer 1

[Arup Ghosh]

I'm not sure but the FindMarkers code suggests it only takes the P value from DESeq2 results when you specify test.use="DESeq2".

[suvratha]

so you mean to say the Log FC reported is natural Log itself like the default?

[ATpoint]

Seems so, the relevant line is here: https://github.com/satijalab/seurat/blob/master/R/differential_expression.R#L953 Only p-value is exported from DESeq2 as Arup Ghosh says so FC is calculated elsewhere, you would need to scan the script for it to be sure. I think this is the relevant FC line https://github.com/satijalab/seurat/blob/master/R/differential_expression.R#L539 so natural log.

[Arup Ghosh]

Line 552-555 have the code for fold change calculation.

[ATpoint]

Yes but that is only the execution, the function declaration starts at 539.

return(log(x = rowMeans(x = expm1(x = x)) + pseudocount.use)) is the relevant part

with ?log:

log computes logarithms, by default natural logarithms,

[Antonio R. Franco]

Take a look to this lane. LogFoldChange is calculated as Log2

https://github.com/satijalab/seurat/blob/master/R/differential_expression.R#L1414

[ATpoint]

What makes you say that? This line is part of a function that is not even exported. Not sure for what it is. Any ideas? It is not (at least not based on this script) related to the DESeq2 function of this script.

Answer 2

[Santosh Anand]

In the last version (v2), everything was done on a natural log scale. See

https://github.com/satijalab/seurat/issues/332

https://github.com/satijalab/seurat/issues/741 (point 1)