I have RNA-seq for two groups of patients: Responders to chemotherapy (n=9) versus non-responders to chemotherapy (n=24)

I have done DESeq2 and obtain fold change of responders versus non-responders in which positive fold change means down-regulation in responders (if I am not wrong)

> head(df)
    gene       FC
1 286499 7.116071
2 260436 6.691189
3   2267 5.751322
4   5968 5.522079
5   4103 5.328008
6 442709 5.036087

I have done GSEA in pre-ranked way

got this plot for KRAS pathway

enter image description here

I interpret this as; KRAS pathway in being down regulated in responders

Am I right?

Please correct me

When I separated genes for up and down regulated

For up regulated genes I did not get any significant GSEA but for down regulated genes in responders I got this

enter image description here

SO does this mean that KRAS pathway is being up regulated in responders because I have used down regulated genes in responders?

  • $\begingroup$ log2(responders/non-responses) > 0 implies responders/non-responders > 1, which means responders > non-responders, which in turn means up-regulation in responders. How did you get to down-regulation in responders? Am I missing something? $\endgroup$ – Ram RS Aug 11 '20 at 15:37
  • $\begingroup$ Because few genes at the start of plot going to be up regulated but at the end more gene set started to be down regulated :( $\endgroup$ – Exhausted Aug 11 '20 at 15:44
  • $\begingroup$ I'm asking a basic math question. You're saying that a group of genes that have more expression in one subset compared to the other subset are down-regulated. Common sense dictates more expression equals up-regulated. Can you please explain that? $\endgroup$ – Ram RS Aug 11 '20 at 16:50
  • $\begingroup$ If a gene is expressed 4 units in responders and 2 in non-responders, log2FC = log2(4/2) = log2(2) = 1, which is >0. Thus, a gene up-regulated in responders (has more expression) will have log2FC > 0. How are you equating log2FC>0 with down-regulation in responders while measuring fold change with responder expression in the numerator? $\endgroup$ – Ram RS Aug 11 '20 at 16:52
  • $\begingroup$ @Ram RS that is why I got confused I know the direction of expression but I can not understand the GSEA part $\endgroup$ – Exhausted Aug 11 '20 at 18:31

In Deseq2, it depends on contrast eg. if your contrast <- c("Condition", firstC, SecondC) is then -ve is downregulated in FirstC and same for +ve. It is always better to take look on count data for the confirmation (responders versus non-responders samples).

for more detail you can refer here

  • 1
    $\begingroup$ Thank you but actual question is about GSEA $\endgroup$ – Exhausted Aug 11 '20 at 7:29
  • 1
    $\begingroup$ The presence of most of the genes at the end of the plot (~4k ranks) is more than expected by chance. So yes, in general, we can say the pathway is downregulated because it has the highest score. ( but as I explained before you need to verify, in which condition those genes are down-regulated), then you can conclude in which state the pathway is down-regulated. You may also like to plot GSEA separately for both the list of statistically significant down-regulated and up-regulated genes. $\endgroup$ – Amarinder Singh Thind Aug 11 '20 at 15:44
  • $\begingroup$ @ Amarinder Singh Thind please look at the edition of my main post $\endgroup$ – Exhausted Aug 11 '20 at 16:10
  • $\begingroup$ your statement is a bit confusing as you wrote, "SO does this mean that KRAS pathway is being up-regulated in responders because I have used down-regulated genes in responders?". @Exhausted If I were in your place, I might take a look at all the genes in your pathway. Then try to investigate how many of them are part of the significantly DE up-regulated list and down-regulated list and try to justify my fining based on the ratio of down/up genes of the pathway. Higher the ratio, more are the chances that pathway in overall down-regulated. $\endgroup$ – Amarinder Singh Thind Aug 12 '20 at 1:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.