ssGSEA score

[Massar Dieng]

Hello,

I am using ssGSEA through Genepattern to analyze RNAseq data. In the output folder, there are three files: gp_execution_log.txt, stdout.txt, and my test file (test.gct). It appears that the .gct file contains pathway scores (in attachment). However, I've noticed that the numbers are relatively large (-134; 234, etc.) compared to scores reported in some publications (e.g., -0.1; 0.2; 0.3). Could you please explain the reason for this discrepancy, and is there a way to adjust the scores obtained?

Thank you,

Mass

[Anthony Castanza]

Hi Mass,

These seem pretty typical for ssGSEA scores, the other scores you mention are more along the lines of what you'd expect to see with unnormalized GSEA enrichment scores, which are a substantively different metric than the calculation ssGSEA performs and are not directly comparable. But that can depend on what kind of normalization you used for your input data. What did you use here? Normalized counts? TPM? FPKM? We generally recommend TPM here, and strongly recommend against any metric that doesn't take into consideration gene length bias. But the specific choice of metric will impact the scale of the scores.
If you really need scores that are comparable to standard GSEA scores, I might suggest trying GSEA preranked. If you only have a small number of samples it might be reasonable to run each of them through a GSEA Preranked run. That said, the scoring method for ssGSEA is better suited towards this sort of analysis.

-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/ba6b77a9-298a-4317-bad8-d00ed3b96433n%40googlegroups.com.

[Massar Dieng]

Hi Anthony,

Thank you for your response. It's reassuring to hear that the scores I obtained are within the expected range.  My input data is in TPM, but I should mention that I also performed "IQR normalization" before inputting it into ssGSEA. Do you think it's more appropriate to use the TPM directly without applying additional normalization? 
All the best,

Mass

To view this discussion on the web visit https://groups.google.com/d/msgid/gsea-help/CAGCeyZxJe-xF%2B3w0eCd-B4SAEz3t89Qq3SPT4NhbByR4bm%3DgWA%40mail.gmail.com.

--

Massar Dieng
Associate Research Scientist – Lab Manager
NYU Abu Dhabi
Office Tel (UAE): +971 2 628 5508
Mobile (UAE): +971 56 998 4054

NYU Abu Dhabi, Saadiyat Campus
P.O. Box 129188
Abu Dhabi, United Arab Emirates

http://nyuad.nyu.edu

[Anthony Castanza]

Hi Mass,

My recommendation is to provide the TPM scaled data. Internally ssGSEA performs a z-score like normalization and this should be sufficient, however I admit that we have not tested this particular combination.

-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/CAG%2BeJK4LbCtiGsHrU-2HXxuSkjAqjg%2BiDF%3DjHSTWkspyY3uAqg%40mail.gmail.com.

[Massar Dieng]

Thank you so much Anthony.

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/-E661FfaPaw/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/CAGCeyZzaUFtEEmD%3D30%3Dz7MGXNaKQS-WakPtxCkY07v%2BTuDF_Mw%40mail.gmail.com.

[Stas Kurpe]

Hi Antony,

Could you explain how method normalization (Z-score or Median scaling) improves usage? Median scaling involves subtracting the median of all scores and dividing by the median absolute deviation (MAD) of all scores. On the other hand, Z-score normalization subtracts the mean and divides by the standard deviation. However, ssGSEA returns non-normally distributed gene signature scores. I think that Median scaling is better suited for use. Am I right?

вторник, 14 ноября 2023 г. в 18:50:10 UTC+4, Anthony Castanza: