Question: only a few differential expressed genes
0
gravatar for A
4.7 years ago by
A3.9k
A3.9k wrote:

hi,

I perform differential expression analysis on my microarray data but only 4 genes found to be expressed at fold change >2 and FDR < 0.01 and only 14 genes in FDR < 0.05

how it possible?? then I have to only count on pvalue?

thank you

R gene • 2.4k views
ADD COMMENTlink modified 4.7 years ago by Carlo Yague5.5k • written 4.7 years ago by A3.9k

Is that a log fold change, or actual fold change?

ADD REPLYlink written 4.7 years ago by andrew.j.skelton736.1k

it is a log2 fold change. by pvalue < 0.05 , 500 genes and pvalue < 0.01, 151 genes at fold change > 2 deferentially expressed. actually i don have replication.I have 4 treatments versus 4 controls which treatment is in 4, 8, 24 and 48 hours. box plot for QC showed the data are very well.

ADD REPLYlink modified 4.7 years ago • written 4.7 years ago by A3.9k
1

If you're looking for a log2 fold change of 2, then you're actually looking for 4 fold differences.

ADD REPLYlink written 4.7 years ago by andrew.j.skelton736.1k
1

Those numbers are more reasonable. If you want a treshold on the fold change, you should use something like :

abs(log2FC)>1

to keep both under-expressed and over-expressed genes (with a ratio of at least two fold).

ADD REPLYlink modified 4.7 years ago • written 4.7 years ago by Carlo Yague5.5k

How big is your dataset? Is it possible that there is no biological difference between your samples? Is it possible your samples are contaminated? Have you QC'd the microarrays?

ADD REPLYlink written 4.7 years ago by russhh5.5k
5
gravatar for Daniel
4.7 years ago by
Daniel3.8k
Cardiff University
Daniel3.8k wrote:

If your data tells you something you didn't expect you can't automatically assume that the data is wrong. Maybe that is what is real.

Alternatively maybe there is human error in the processing, or stricter parameters than are required. Fold-change < log2 can still be interesting, there's no real reason to cut off there.

FDR of < 0.01 is stricter than I would usually use (< 0.05), but you can't just ignore FDR (That's like saying I don't care if I have false positives, as long as I get positives).

But maybe what you have is the real biological difference. Here is my new favourite picture to illustrate the point:

enter image description here

ADD COMMENTlink modified 4.7 years ago • written 4.7 years ago by Daniel3.8k
1
gravatar for Benn
4.7 years ago by
Benn8.0k
Netherlands
Benn8.0k wrote:

I don't know what groups you compared? Did you expect a lot to change or not?

If you want to loosen up, try a FDR < 0.05 (don't use uncorrected p-values), and get rid of the fold change cut off.

ADD COMMENTlink written 4.7 years ago by Benn8.0k
1
gravatar for Carlo Yague
4.7 years ago by
Carlo Yague5.5k
Canada
Carlo Yague5.5k wrote:

I think a volcano plot would be useful to see if the low number of DEG comes from the FDR cutoff or fold change cutoff or both.

To add to the other suggestions, low number of DEG could be caused by very bad replicates (or worse, wrongly associated replicates).

ADD COMMENTlink written 4.7 years ago by Carlo Yague5.5k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1132 users visited in the last hour