"logfc" parameter in findmarkergene(), log2 or log10?

[Leorio183]

Anyone tell me?

[multitalk]

Hello. It represents log1p in Seurat

[Leorio183]

sorry, I still can't understand, I know in Seurat, there is a parameter "base" in FindAllMarkers(), with this "base", I know the logfc means log base (FC), and the base = 2 in default.

However, no information about the base in logfc parameter in findmarkergene() in scCATCH.

This makes me really frustrated. I just mean, A compare to B, if A/B=2, then the logfc = ? in the function.

Thank you very much!

[Leorio183]

I am very sorry, but i really can't understand. I don't know the base in the logfc calculation.

[Leorio183]

I know log1p(x) means loge(1+x), also known as ln (1+x). This means when logfc is set to 0.25 by default, then we get:
ln (1+x) = 0.25, where x is the fold change.
Then Foldchange = 0.28?

We compare to seurat, when we set logfc = 1, we get Foldchange = 2^1 = 2.

So i am very confused about the default setting logfc = 0.25, this means it is not actually select up-regulated marker, since the Foldchange is 0.28

[ShanshanHe-44]

I think it depends on which data you used as input.

As the default/expected input here is normalized expression matrix(Seurat NormalizeData()), which is `xnorm <- log1p(x = x / sum(x) * scale.factor) from Seurat.

In scCATCH, logfc is computed as:
The default logfc =0.25 here based on NormalizeData(), which is（mean(log1p(expr_A)) - mean(log1p(expr_B))） ≥ 0.25, means ~1.28x upregulation

scCATCH/R/utils.R

Line 148 in 06c6ffb

clu_marker2$logfc <- as.numeric(apply(ndata1, 1, mean)) - as.numeric(apply(ndata2, 1, mean))

This is just a difference in expression values, based on the input normalized expression matrix. scCATCH does not further apply any log operation when computing logfc.