What could be a good balance between baseMean and Log2FoldChange?

Question

[Apex92]

Dear all,

I have list of differentially expressed miRNAs (DESeq2 output) and want to choose about 10 important candidates for downstream analysis (GO analysis with targets of selected miRNAs).

I was wondering what could be a good balance in selecting these miRNAs considering both baseMean and the Log2FoldChange?

Is it a good idea to first rank the differentially expressed miRNAs based on the Log2FoldChange and then select the 10 candidates based on having highest baseMean?

What could be your suggestions?

Thank you

Answer 1

[Trivas]

There's no right answer for this. Top 10 by l2FC is reasonable, so would be selecting the top 10 with the lowest p-adj values. I tend to ignore basemean for filtering but I would love for someone to give me a convincing argument for why I should pay closer attention to it.

[Apex92]

Thank you @Trivas for your answer - The thing is that, with baseMean over 100 because I feel that they have most likely regulatory impact. This is of course slightly a philosophical topic but imagine if a DE miRNA has a big l2FC and very low padj, but then baseMean is around 20, would you consider this candidate to have a robust effect in a cell by itself? Or insted the same situation but with a baseMean over 100?

Answer 2

[Apex92]

After lots of brain storming I though the best way is to make a scatter plot from the DE candidates with having x-axis as L2FC, y-axis as baseMean and dot colors based on p-adjusted value (like padj > 0.01 gray, padj <= 0.01 red and padj <= 0.0001 orange).

Then based on this plot I could decide which ones to take for downstream analyses.

[seidel]

It's more typical to reverse your axes, and put the log2FC on the y-axis, and the log2(basemean) on the x-axis. This is called an MA plot, and is a common way to look at this sort of data. Of course adjusted p-value is the way to select candidates if you can, but if you want to rank your candidates taking basemean into account you can take a selection and rank them by absolute value of log2FC * log2(basemean), as this will give some weight to those with higher expression.

[Apex92]

Thank you seidel for your comment. - I mistakenly swapped the x and y-axis in my comment above ( I actually had put baseMean on the x-axis and L2FC in the y-axis).