Differences between lfc=log2(1.5) and lcf=1.5

Question

[Joe Kherery]

Dears,

What is the difference between lfc=log2(1.5) and lcf=1.5 ?

topTable <- topTable(fit, coef=1, number=Inf, adjust.method="BH", **lfc=1.5**)

Which is the correct one to use and why?

Another question is what cut-off to use???

I see many articles using |log2fold change (FC)|≥1.5 and others using |log2fold change (FC)|≥1.0? with a Q-value < 0.05.

If you can tell me a reference about it, I will be grateful

Regards,

Answer 1

[h.mon]

What is the difference between lfc=log2(1.5) and lcf=1.5 ?

If you are asking about which one to use in topTable, the correct is topTable( ..., lfc = 1.5 ), as the fold-changes reported (and used for thresholding) are already log2(FC). However, topTable should not be used, and topTreat is to be preferred - read the topTreat manual page with ?topTreat.

As for "correct" threshold, my view is that there isn't one. It is of course necessary to control for false-positive discovery, hence the q-value of 0.05 (or lower). But for fold-changes, I guess it depends on if one is interested in all changes, or only big changes in expression.

One of the benefits of using log2(FC) is that the values are intuitive: a log2(FC) of one means one treatment has double the expression of the other; a log2(FC) of 2 means one treatment is 4 times more expressed than the other, and so on. However, using log2(FC) = 1.5 means one treatment is approximately 2.828428 times more expressed than the other - as for arbitrary thresholds go, you can't go much more arbitrary than this.