How to understand the Gene markers in the comparison?

[于霆峰]

Dear  GSEA team,

Based on my RNA-seq results, I performed a GSEA . Can I interpret the top 50 features in the heatmap and the ranked gene list correlation plot as primarily reflecting the comparison between AP and HC? Additionally, does the ranked gene list correlate with AP and HC, such that genes at the top of the list show a stronger association with either AP or HC?

Thank you so much !!!

Jason

[Anthony Castanza]

Hi Jason,

I'm not 100% sure I understand what you're asking here.
The ranked list is computed with respect to the phenotypes you selected in your GSEA setup.
In an A vs B comparison, the ranking computation will be performed as A with respect to B such that if a gene is more highly expressed in A than B then it's ranking metric will be positive, and if it is more highly expressed in B than in A it's metric will be negative.
The GSEA algorithm (weighted K-S) is then used to determine if, on balance, the genes in a given set are more positive or negative, and thus enriched in A (on balance positive), or in B (on balance negative).
I wouldn't recommend using the top features of the heatmap directly for any specific analysis, you would want to run a standard differential expression analysis pipeline to compute per-gene significance statistics, which GSEA does not do.
We mostly provide these heatmaps and correlation statistics for data quality control assessment.

Does this answer your question?

-Anthony

Anthony S. Castanza, PhD

Curator, Molecular Signatures Database
Mesirov Lab, Department of Medicine

University of California, San Diego

[于霆峰]

Thank you for your quick response, Anthony! 

Your answer was so helpful and I would like to ask an additional question: So the ranking metric in the Gene markers' part is a quality check？Higher score means higer reliability of the defferential expression?   

Thanks again—sorry for all the questions!

Best,

Jason

[Anthony Castanza]

Hi Jason,

The ranking metric is just GSEA's internal computation of the differential expression and is used in computing the weighted K-S test. We don't compute per-gene significance statistics, so if you're interested in per-gene metrics (e.g. for the reliability of the differential expression), you'd need to use a tool that is designed for that purpose (e.g. DESeq2). You would expect to see a broadly similar result, but again, different tests and different statistics for different purposes.

-Anthony

Anthony S. Castanza, PhD

Curator, Molecular Signatures Database
Mesirov Lab, Department of Medicine

University of California, San Diego