input of run_de function when using a Seurat object

[pandaqiuqiu]

Thanks for your useful tools for scRNA analysis! I just want to know for sure about input of run_de function when using a Seurat object:

Is input raw counts or "RNA" assay when de_family = "singlecell"? Is input only raw counts when de_family = "pseudobulk"? What about "mixedmodel"?

I appreciate it if you answer this question.

[jordansquair]

By default it will use "RNA" as the assay. You can always pass the count matrix in directly.

Yes of course pseudobulk and mixedmodel use counts only.

[pandaqiuqiu]

Make sense! Thanks so much. Another question is, If I set label as input of label_col like this: label = factor(label, levels=c("A", "B")), and then run: run_de(expr, meta, label_col = label). Does the avg_logFC of the result mean A vs B (as reference) or not?

[jordansquair]

The first level of the factor would be used as the reference. so levels = c("A", "B") with a logFC > 1 means B > A.

[pandaqiuqiu]

The first level of the factor would be used as the reference. so levels = c("A", "B") with a logFC > 1 means B > A.

Thanks for pointing out my mistake. I really appreciate that.

[pandaqiuqiu]

I just encountered something confusing. As you have described, I used the following to calculate B vs A (A as reference), with code and result as below.

seurat$TimePoint <- factor(seurat$TimePoint,levels = c("A","B")) DE = run_de(seurat, cell_type_col = "Celltype", replicate_col="Sample", label_col = "TimePoint",
de_family = "pseudobulk",
de_method = "DESeq2",
de_type = "LRT",
n_threads = 4)

But when I used the same data as input and same method(deseq2+LRT) following as tutorial (https://github.com/hbctraining/scRNA-seq_online/blob/master/lessons/pseudobulk_DESeq2_scrnaseq.md#scatterplot-of-normalized-expression-of-top-20-most-significant-genes), the logFC result is reverse, the key code and results is like this:

dds$group_id = relevel(dds$group_id, ref = "A") dds <- DESeq(dds, test="LRT", reduced=~1)

Can you take a look to see what's causing the problem?

[jordansquair]

I'm not exactly sure from your post what the problem is. You can just plot some of these genes to see if the direction is correct no?

[pandaqiuqiu]

Yes. I used to_pseudobulk() function to get counts at sample-level for plotting A1BG gene. The stat and plot like this:

Also, I used plotCounts(dds) function to get normalized data at sample-level for plotting A1BG gene. The stat and plot like this:

It seems A1BG is down-regulated in B compared A, so log2FC should be negative in result. What do you think about it?

[jordansquair]

Yes sorry - typo above -- the second factor level is used as reference.

[pandaqiuqiu]

Yes sorry - typo above -- the second factor level is used as reference.

Thanks. Good to have this figured out.

[vkartha]

Follow up to this - Is it always that the second factor level is used as a reference? What if it is a character (and not a factor) and is unspecified, will the run_de function convert to a factor and still use the second one as reference (and will this be alphabetically sorted to determine what is second?)? I find this to be a bit confusing since most tools like DESeq etc use the first level as a reference