RNA-Seq experiment has low fold change

Question

[firestar]

My differentially expressed genes (~600 DEGs) have low fold change (<1). Is this normal and what does it mean?

More details: I work with Zebrafish and had 2 conditions and 12 samples per treatment (24 samples total). I have found around 246-931 differentially expressed genes out of the 23000 or so ensembl annotated genes between my 2 conditions. I say 500-700 because I have tried DeSeq2 (931 DEGs), EdgeR (549 DEGs) and Limma-Voom (246). These genes are FDR controlled at 0.05 and filters like cooks cutoff and independent filtering etc have been applied if applicable. No filtering was applied on fold change. These genes were found to be enriched in 30-40 or so GO terms and 3 KEGG pathways. The GO terms were too vague (cell metabolism, cytoskeletal protein binding etc) to say if they actually mean anything in relation to my treatment. I am a bit concerned about the low fold change. Does this mean that the DEGs are false positives? Or does it mean that the DEGs may be significant and positive but the biological effect is low?

[russhh]

It all sounds fine to me. I'm guessing your RNA source is a pool of a whole bunch of cell populations, so a large fold-change in a subpopulation will get watered-down.

Answer 1

[WouterDeCoster]

Fold change doesn't necessarily reflect biological effect. E.g. a transcription factor with small increase in expression might have a drastic effect. Could you elaborate on your experimental setup? Which treatment are you comparing in how many samples?

[firestar]

2 conditions x 4 families x 3 replicates = 24 samples

my design was as follows: ~family+condition

Answer 2

[Andrea]

I agree with WouterDeCoster. Larger fold change certainly builds your confidence that a given gene is a true positive but it does not reflect the significance if its change between your two conditions. Consider the number of replicates and the internal variability between them within each group. These factors will affect the fold change. The DEGs that you have found with the three different methods, do they overlap? If they do, that should be a good indication of a true positive result. My guess is that there should, at least, be a decent agreement between DESeq2 and EdgeR. Also, note that if you are looking at shrunken fold changes from DESeq2, they will be slightly smaller in magnitude than those from EdgeR.

As of GO terms, depending on what you used for that, it’s also entirely normal that the analysis gives vague and redundant results. If you haven’t, try RAMONA (http://mips.helmholtz-muenchen.de/mona/). It’s a Bayesian approach to GO annotation. It will produce very few, if any, significant GO terms but they should be relatively specific and non-redundant. Just make sure that you select a “single” model there as it’s designed for multiple omics levels.

[firestar]

Thanks for the comment. From the DEGs I mentioned in my question [DeSeq2 (931 DEGs), EdgeR (549 DEGs) and Limma-Voom (246)], 188 DEGs are shared/common between them.

[firestar]

The RAMONA tool does not seem to support Danio rerio (Zebrafish).