references for fold-change cutoff

Question

[CrazyB]

To my knowledge,

there is no consensus cutoff value for the fold-change in transcriptomic data (correct me if this is wrong). However, one reviewer for a submitted manuscript asked that we provide references to support our cutoffs for significance selected. I wonder if there are "good" references that either spell out a "hard" cutoff values for fold-change or state clearly that there is no consensus cutoff value for the fold-change.

Any suggestion ? Thanks

[CrazyB]

Our decision tree is as follows -

1. calculate the fold-change with FDR
2. remove all genes that have FDR >=0.05
3. rank the remaining genes based on fold-changes
4. make an arbitrary cutoff of the fold-change - we set it at a low end (1.2-fold)
5. pool all genes that fulfill the FDR and fold-change and do the gene set/pathway enrichment analysis

The biggest issue is references are needed to justify whatever fold-change we set to include the genes for pathway enrichment analysis. Thus, I am hoping either

(a) a reference says there is NO hard cutoff

or

(b) a reference says there is a hard cutoff, but with caveats

Unfortunately, whether it is "you think", "I think" or "someone else on this forum thinks", none will satisfy the reviewer.

Any suggestion?

[russhh]

I don't think you need a reference. What you need is to show that your results are robust to the arbitrary cutoffs that you chose. Iterate over a range of FDR-thresholds and log-fold-change thresholds, rerun your pathway analysis, keep those genesets that are signficant regardless of your arbitrary choices.

[Fabio Marroni]

Did you base the decision of differential expression only on fold change (which is in my opinion wrong) or are you using pvalue and then filtering out genes with low fold change (which is in my opinion acceptable)? After this, I provide you my personal opinion (for what is worth!) You are right, there is no consensus for cutoff value for the fold-change in transcriptomic data. This is because, theoretically, the decision on which genes are differentially expressed should be based on the (adjusted) pvalue, not on the fold change. So, in my opinion there is no reference and no excuse to justify the use of a fold change to decide DE genes. However, I agree that, if in an experiment you find some genes that have a significant p-value for DE, and some of them have very low fold changes, then it is possible (and desirable) to decide that significant genes with very small fold changes have a statistical significance but not a biological significance (do you care about a change in expression of 5%?).

Answer 1

[CrazyB]

[cancelled by OP - should be a comment not an answer]

Answer 2

[Fabio Marroni]

Maybe this or this might help?