GSEA and genes with 0 TPM's

Question

[didemdkn]

Hello everyone,

I have a very basic question which I do not know how to reason it. I conducted a RNA-Seq experiment without replicates (please do not blame me :() and i calculated the fold changes between two conditions (let's say X and Y) using TPM values. Now, I want to find some enriched sets using GSEA. To do that, I need to choose the highest fc values or prerank the fc's.

Here is the question,

Some of the genes have 0 TPM's in condition X and more than 0 in condition Y. I added 0.001 to all the TPM values to calculate fold changes to get rid of the 0 values. However, those genes are looking like they have 8976976 fc (just a huge number). However, they reason they do not have any TPM in one condition can be low coverage, or they were not there in the condition X and started to be expressed in condition Y. How should I treat them while appliying GSEA? Should I omit them or put another filter for them etc.?

Any help is appreciated. Thank you for your help!

[Sandeep]

What strategy did you finally apply?

Answer 1

[Soheil]

When we have replicates we usually filter out all the genes with low counts (say less than 10) across all samples but here it wouldn't really make sense. As you figured yourself, without any replicates you can't really tell if a gene or pathway is up/down-regulated. To be more precise, without replicates it is not clear whether the difference between test and control is just by chance or statistically significant. Finally, I think a single-sample enrichment analysis (ssGSEA or ssGSVA) would be a better option here. Yet you won't be able to tell if the difference observed between control and test is statistically significant.