Running GSEAPreranked using RT-qPCR Data

[Patrick Tamukong]

Dear GSEA Support, 

THANK YOU on responses to my previous questions. I am working with a small dataset of normalized RT-qPCR data on only 420 genes. I have used the 2^(-deltadeltaCT) Method to compute fold changes for my genes and then used those fold changes as the ranking metric for running GSEAPreranked. While I found some pathways, the FDR q-values are disappointing (all greater than 0.05 even for pathways with nominal P-values less than 0.05). My enrichment plots are not smooth and I suspect this could be due to my small sample size. I have a few questions:

1) Is my approach correct and is it okay to run GSEAPreranked with fewer than 450 genes?

2) Are pathways with nominal P-value<0.05 significant if their FDR q-values are large (larger than 0.05)?

3) Can I run a cox regression analysis of my genes and use the cox P-values as the ranking metric for running GSEAPreranked? I know the ranking metric needs to be unique for the gene list and concerned that a cox regression may give same p-values for some of the genes, complicating a GSEAPreranked analysis. 

4) Am I correct to believe my enrichment plots aren't smooth due to the few genes I have?

Thanks so much on your help. 

Patrick

[Anthony Castanza]

Hi Patrick,

 

I would not think that 420 genes would be enough to reasonably run GSEA. This seems to be borne out in your description of the plots. I would strongly doubt any conclusions drawn from such an approach.

In particular, there is potentially a huge gene selection bias here. Where the 420 genes randomly selected? Or is this a targeted panel? I'm not sure just the standard housekeeping controls would be sufficient to get a reasonable background.

 

Tied values in the ranking metric will only add an element of arbitrary ordering to an already extremely limited dataset and what would normally be a minor effect on the enrichment score could be substantially exacerbated.

Yes, the jaggedness you're observing in the ranking is likely from the extremely limited gene lists.

 

I really don't recommend this approach.

 

-Anthony

 

Anthony S. Castanza, PhD

Curator, Molecular Signatures Database

Mesirov Lab, Department of Medicine

University of California, San Diego

--
You received this message because you are subscribed to the Google Groups "gsea-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gsea-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/e2d1c1fc-103d-448a-9060-e62439d989can%40googlegroups.com.

[Patrick Tamukong]

Thanks Anthony. 

Suppose I had RT-qPCR data on many genes (say 20,000), would my approach of normalizing the data and using 2^(-deltadeltaCT) to find fold changes, then using those fold changes as the ranking metric in GSEAPreranked be correct? I am asking this just in case we choose to obtain RT-qPCR data on more genes and find enriched pathways. 

Thanks on your time. 

Patrick Tamukong

Cedars-Sinai Medical Center

Los Angeles CA 90048

"It is also a good rule not to put too much confidence in experimental results until they have been confirmed by theory"-- Sir Arthur Eddington

You received this message because you are subscribed to a topic in the Google Groups "gsea-help" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/gsea-help/ffyYQVYBH8U/unsubscribe.
To unsubscribe from this group and all its topics, send an email to gsea-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/SJ0PR05MB7609ABBDF147B26B95F9EF1FF7B09%40SJ0PR05MB7609.namprd05.prod.outlook.com.

[Anthony Castanza]

I think that would probably work, you'd want to look at the Log2(FC) distribution to see if it follows a relatively symmetrical curve. That said, at that scale you should just sequence the sample as the cost-benefit of sequencing far outweighs any advantages RT-qPCR might have.

To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/CAGFyC%3DRSWV7ocd58T-EM32eC2WVG4k9_Mi6UVrEoS0Qa0k9iWg%40mail.gmail.com.

[Patrick Tamukong]

Oh yes, thanks Anthony. 

Much appreciated. 

Patrick