Hi Sonti,

yes and no :)

Yes... Taking +/- 2 L2FC as another cutoff will make you loose some (important) genes. Typically, ppl take +/- 1 L2FC as further cutoff.

2.: based on your L2FC, regardless of the threshold applied, you can or should split to either positive or negative values --> representing either UP or DOWNregulated genes between your groups.

--> do also pathway analysis with just UP or DOWN reg genes (don't mix both groups... :) )

In the end, its very subjective on which cutoffs to apply.

I would start with FDR (as you did) AND +/- L2FC.

Hope this helps a bit :)

Best,

Michi

Hi Sonti Roy,

Basically, there is no one-size-fits-all criteria to filter DEGs for downstream functional enrichment analysis. However, as Micheal Schulz has mentioned, generally people use Log2FC > 1 or < -1 (which means that the gene expression is at least 2-fold higher/lower than control) to filter their DEG list. Log2FC > 2 or < -2 may be too stringent, leaving you very few genes for downstream analysis.

However, I do not agree to split UP- and DOWN-regulated genes for pathway analysis. This is because upregulation doesn't necessarily imply the activation of a pathway as a highly expressed protein could act as an inhibitory molecule to a certain pathway. Likewise, downregulation may not indicate inhibition of a pathway.

Normally, I subject the entire DEG list (after applying cutoffs) to pathway analysis. Based on the results of pathway analysis, I will map DEGs to selected pathways to visualize the gene-gene interaction. I find Pathview ( https://pathview.uncc.edu/analysis) very useful for this purpose.

I hope this helps.

Regards

Hong Sheng

Hi Roy,

I am agree with Michael and Cheng answers. You have both options for pathway analysis (split up/down regulated genes or don't split them). In split case, you can explain them separately as up or down regulated genes in specific pathways; while in combine you can explain them in general. Its upto you, how you want to carry out your presentation style.

Thanks Michael Schulz , Hong Sheng Cheng .

Thanks for sharing pathview Hong Sheng Cheng, I have one more question regarding RNA Seq Analysis.

I have done analysis using CLC genomics and in Galaxy using Tophat and cufflink. In both analyses, I am getting P-value as Zero in many genes so I change it to 1E-16 and made a volcano plot Using Log2 Fold change and -log of FDR adjusted P-value. The Genes having P-value 1E-16 are coming at same place, but I have never seen any volcano plot like this in any publication. So what should I do with genes having p-value as zero?

Hi Sonti Roy,

I do not have prior experience with CLC genomics. However, based on this site ( https://secure.clcbio.com/helpspot/index.php?pg=kb.page&id=448 ), p-value = 0 is because of very small p-value which gets rounded off to 0.

If you use technical replicates for the RNA-seq, you may get very small variance within a treatment group, resulting in many genes with very small p-values. In this case, there is nothing unusual to include the genes in your downstream analysis.

Hi Sonti, fold change threshold might remove some important genes. Especially regulatory genes like TFs that may have nominal changes but induce a large effect across the network of genes.

Chirag