Check this link: https://www.biostars.org/p/342756/

I think you have a conceptual misconception

Log2 fold changes are fairly straight forward as explained in the link provided by Miguel. The real issue is as to how the readset alignments to the transcribed gene regions were normalised and the consequent confidence you should have in the reported fold changes. Lets assume that your company doing the DE analysis has normalised for technical replicates, readset sizes, and number of read alignments. But have they provided you with a P-value or some other measurement of confidence in the claimed DE foldchanges and what documentation was provided explaining their analysis methodology?

My personal view is that if coverage is less than 1000 reads in either sample then, irrespective of any P-value threshold, it is worth while checking the read coverage in transcribed genic regions of interest for evidence of differential PCR amplification. Some transcript fragments may be PCR differentially amplified many thousands of times between samples dependent on the number of PCR rounds. End result is you observe a transcribed gene region with 100 reads in one sample and 2000 reads in the other sample, nominally 200x DE, yet in the second sample 1900 reads all aligned to exactly the same loci!

If you do not have the time or resources to check for coverage along the transcribed gene regions then at the very least be suspicious of genic regions having less than a 2x change.

I keep referencing transcribed gene regions as opposed to genes, why? It is because referencing read counts attributed to genes takes no account of the differential {multiple} isoforms that a specific gene may be being expressed between the samples. Referencing transcribed regions forces the researcher to consider the possibility of differential isoforms rather than just thinking in terms of a simplified gene.

Yes, I know I have diverged from your original question but I do think it is important to encourage researchers to more critically think about what DE fold changes in terms of alignment counts really represent and not just blindly report them as DE fold changes. I won't further diverge and mention all the other considerations such as alternative start sites, transcript half live, circularisation, RNA editing, ncRNA in retained introns or …. . Just remember that one mRNA does not equal one translated polypeptide!!!

Why are you worried with the close mean and the low SD? Rather,I will say your data is consistent,so report accordingly.

Ajit kumar Roy - consistent with what?

In order to have any confidence in a SD then many samples are required relative to the population otherwise the confidence interval will be rather large. With no prior information about the population distribution then you should target having at least 30 sampled data points to be up around the 95% confidence interval. Usually in DE experiments we may have multiple technical replicates and multiple sample replicates for both the control and experiment. But you would be very fortunate to have more than two or three technical and/or sample replicates. So attempting to calculate a population SD from so few samples is rather meaningless in terms of resultant confidence.

It is further complicated because when sequencing you are actually finitely sampling from a mixture of various fragmented transcripts. If a mixture component is proportionally increased then other components have a lower probability of being sampled simply because a finite number of samples are being taken from the mixture. Thus it is feasible for a component (transcript) to have apparently decreased in abundance simply because some other transcript is now present in higher abundance in the mixture. Without mixture count normalisation then DE reported fold changes (or lack thereof) can be very misleading.

It is very informative to run DE over a couple of technical replicates from the same biological sample – you may be surprised at the number of transcribed regions showing as having DE count fold changes.

I've said enough and hopefully sufficiently muddled the water!

Thank you all for the information, but my data is in log2 and not log2 FC. Meaning my values are from 5.40 to 22.8. I don’t have values that are 0.5 as mentioned in the link provided. I think that is why I am a bit confused. I have set up a call with the company that did the analysis to try and understand better the output.