What is the safe fold change to consider in a RNA-seq experiment? Fold change > 1.5, FDR < 0.05, P-value < 0.05 and 'Test status' = OK is one criteria which was taken, but I have also seen people considering fold change > 2. I took 3 replicates for the mutant and the wild type each. During microarray times, it was mostly the case that people used to take fold change > 2. Is it safe to take fold change >1.5 for RNA-seq or is it a matter of choice? In my experiment out of 3300 Arabidopsis differentially expressed genes (>1600, about half) lie in the range of fold change 1.5 to 2 and if I take 2 as a cut-off, those genes will be missed out. Is 1.5 fold a genuine expression fold change and should be considered? What is the trend, I was going through a few papers where people have taken 2 fold in RNA-seq too, please advise?
Thanks!
I hate to sound cliche, but there is no one-size-fits-all answer here. However, I would say that simply labeling genes as differentially expressed (DE) based on fold change alone can be dangerous. For example, gene A may go from 1 +/- 1 (95% CI) RPKM (or whatever normalized read count you prefer) to 3 +/- 2 (95% CI) RPKM in another sample group, while gene B goes from 1000 +/- 50 (95% CI) RPKM to 1400 +/- 80 (95% CI) RPKM. Obvious gene A is up 3 fold while gene B is only up 1.4 fold, but given the overlapping confidence intervals of gene A, the change in gene B appears far more significant and could very likely have a more relevant biological impact. That said, with enough statistical power you could have a situation where statistical analyses are yielding DE genes that correspond to a relatively low fold change. In such a case you may want to add a fold change threshold to focus in on genes with a larger change. In general, I'd say that as long as you have the sample numbers to allow for statistical analysis (or even just paired samples that allow for a paired T test), you should include a P value or FDR threshold in designating DE genes. I personally like the EdgeR package that is part of Bioconductor. If your study is too under powered to detect any significant DE genes, you could simply apply a fold change threshold, but confidence in the true DE status of those genes would be extremely low without independent validation (like with RT-qPCR) that yields a statistically significant result.
I hope that helps. Good luck!
Let me add only that if you use the FDR or p-value for selecting the differentially expressed genes, you are using a statistical criterion, while there is no reason to cut at some fold change. The p-value suggest how much extreme the expression value of a gene in the treated sample with respect to its variability in the control libraries, and therefore it is a measure of how affected the gene was by the treatment. Filtering at some fold change is something that can be done for reducing the number of deg in case you get a lot, but remember that there is no reason to think that for different genes the larger the fold change imply a larger effect on the downstream processes. It depends on a lot of additional things. Think about a protein that needs activation by phosphorylation. A small fold change, if it co-occur with the respective kinase, would have a much larger impact than a larger fold change when the kinase is not expressed. This to say that I prefer to be less stringent in detecting the degs, but then I put more care in extracting the important and coherent biological signals from the output e.g. through enrichment analysis.
Hello Sachin,
Having 3 replicates is quite nice. It is reasonable to report number transcripts at both 1.5 fold change and 2 fold change with FDR and p-value in all 3 experiments. It's likely you may transcripts with 2+, 2+, 1.9 that are significant and so close to all 3 replicates above an arbitrary threshold that those values would be included in the 1.5-fold change group and a shame to filter out. If you report the values in a supplementary table, you could annotate or mark the transcripts >=2-fold in the entire >=1.5 fold set with your statistical values. More highly differentially expressed transcripts can be described as >= 4,5,6-fold or whatever if there is interesting biology associated. If you do any enrichment analysis, I would use the 1.5-fold set to retain more members of some class but you may consider a resulting group more significant if the value of the group generally lie at the upper range over a group that may just make it in the lower range. Judgement about the biological relevance of any set of differentially expressed transcripts is a major factor as pointed out above for a kinase and its substrate. Keep in mind some transcripts may be negatively correlated.
Good luck!
Diane
Dear all,
I personally prefer having a filter based on a strong adjusted p-value or FDR (given the fact that most of us use 3 replicates for a sample) and then select based on the biological annotation of the genes (if relevant to the biological question) followed by fold change. My view is that the upstream regulators of a signaling pathway may not have large changes associated with them as compared to downstream genes which are most probably targeted by multiple pathways. Keeping this view I would regard a gene if it seems biologically interesting to my question even if it's FC is equal to or greater than 1.5.
Regards,
Debatosh Das.
can any one tell me about RNA seq data which value people set for DEG up and down regulated ? how people set the specific number of log 2 ratio, rich ratio , FPKM/RPKM , P value , Q value, any specified value for this ? please little guide me. the FDR value is default set by company 0.01 and Q value in some papers used is 0.05.
Calculated false-discovery rate is always higher for higher transcript levels. Similarly calculated fold expression is higher in genes with lower transcript levels. To overcome this problem, I would suggest to read NOISeq approach and follow-up methods.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3227109/
https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-018-5390-6
https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1006792
I did the same question during my PhD. To solve it, I found the differentially expressed genes (DEGs), first using the t-test (p-value), and then using the FDR (q-values). As there is not a "true" threshold to define up and downregulation, we calculated the DEGs fold change, and then we set the threshold at 1 +/- 0.2. It means that, for ratios equal to or less than 0.8, genes were considered as downregulated, while ratios equal to or greater than 1.2 were considered as upregulated. The logic for this was simple: 10% of "error" is normal in an experiment, but 30% is too much; therefore, a safe threshold would be 20%, in this case, for the DEGs ratios. The fold change group between 0.8-1.2 was considered as continuum group (genes that do not change the expression status), and all the 3-groups were analyzed, comparing the p- and q-values. In the end, the pathways enrichment were similar, regardless of the p- or q-values. So, if you are going to find the pathways enrichment, the results would be similar.
In most of the microarray data why only >2 value is significantly considered upregulation? why not1.5 or any other?
- Badges
- Science method