ssGSEA input data

[Katia S]

I am trying to run a ssGSEA analysis but I am a bit confused about the input. Can I input normalised counts as in the GSEA?

Also is it allowed to have zero values?

Many thanks,

Katia

[Anthony Castanza]

Hi Katia,

For ssGSEA we'd recommend a metric that has a gene length normalization built in, such as gene level TPM, or FPKM, since ssGSEA compares within a sample for relative enrichment. Normalize counts is more appropriate for standard GSEA which ranks per-gene differentially between samples.

Zero values are allowed but as they are considered a "tie" in the ranking, when the gene list is sorted, the order ends up arbitrary. This can have a variable effect on the score depending on how extensive the zeros are. The issue here is largest with single-cell data where we don't necessarily recommend ssGSEA for individual cell-level quantitation, but binning to cluster level seems to work reasonably well, as does bulk sequencing data.

-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/8a7a3b20-f419-4532-aafa-303439377781n%40googlegroups.com.

[Katia S]

Hi Anthony,

Many thanks for your reply. That was very helpful!

Best,

Katia

[Katia S]

Hi Anthony,

Can you please let me know based on what do I choose the sample normalization method during the ssGSEA. 

Just to confirm I didn't pre-rank my genes, I just input normalised counts, is that correct?

Many thanks,

Katia

[Anthony Castanza]

Hi Katia,

Like I said before TPM or FPKM are more appropriate for ssGSEA than "normalized counts", at least, the way we normally understand normalized counts is non-length corrected counts normalized by something like quantile normalization, or DESeq2's median of ratios. For ssGSEA it is strongly recommended to use an expression statistic that is length normalized.
The "sample normalization method" parameter should generally be left as "none" it isn't applicable to most data sets and the options will not give a desired behavior anyway if the input data type is not appropriate for the method.

-Anthony

Anthony S. Castanza, PhD

Curator, Molecular Signatures Database

Mesirov Lab, Department of Medicine

University of California, San Diego

To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/3d9924cd-4a99-4473-9d55-3ea13ac2fb21n%40googlegroups.com.