log2 fold change in RNA-seq analysis
2
0
Entering edit mode
7 months ago
wmsalsah • 0

HI everyone, In RNA-seq analysis. for Desq2 I have chosen adj-pvalue <0.05 and log2 fold change > 1 to find differential expressed genes. My question is when I choose log2 fold change > 1 will give up regulated genes, so what about the down regulated genes? Do I need to specify log2 fold change < -1 ?!

fold log2 change • 932 views
3
Entering edit mode
7 months ago

It depends on how you are specifying the log2FoldChange. If you are passing this to the "lfcThreshold" argument of the DESeq2 results function, then no, you don't need to explicityly pass -1 as well, because results uses this to do |log(foldchange)| > 1.

However, if you are manually filtering the output table, then yes, you will need to look for both log2FoldChange > 1 and log2FoldChange < -1, or simply abs(log2FoldChange) > 1.

0
Entering edit mode

Thank you i.sudbery, its very helpful. I used abs(log2FoldChange) from Desq2. I just wonder, why abs(log2FoldChange) >1 not 2 for example?

2
Entering edit mode

A log2FoldChange of 2 means a fold change of 4 because Log2(4) = 2. That means you're testing if a gene's expression is 4 times greater in your treatment compared to your control.

A log2FoldChange of 1 means a fold change of 2 because Log2(2) = 1. That means you're testing if a gene's expression is 2 times greater in your treatment compared to your control.

Same thing in the other direction: A log2FoldChange of -2 means a fold change of 1/4 because log2(1/4) = -2, therefore you're testing if a gene's expression in your treatment is 1/4th that of your control.

With that in mind, whether you want you to test a 4 fold change or a 2 fold change is up to you.

0
Entering edit mode

Yep, it really is just a judgement call, balancing what you know about the biological system, with what you want to do with the resulting gene list and how many genes you get at each threshold.

2
Entering edit mode
6 months ago

As an addendum to Ian's excellent answer, you should really utilize the lfcThreshold argument if you want to do this rather than post hoc filtering the results, as the latter is essentially rendering your p-values meaningless by changing the effect size cutoff. Utilizing the lfcThreshold argument gracefully takes this into account, altering what's actually being tested and adjusting p-values accordingly.

So in short, using it changes the question (null hypothesis) being tested from "Are these genes significantly differentially expressed, assuming a difference of 0 between groups?" to "Are these genes significantly differentially expressed, beyond a two-fold difference between groups?"