heatmapEnrichment() with escape.ssGSEA+scaling vs escape.ssGSEA_normalized

[ayin0510]

Hi Nick,

I hope this finds you well! I have a couple questions regarding the use of the non-normalized vs normalized enrichment scores:

1. What is the difference between using "escape.ssGSEA" with z-scaling as opposed to "escape.ssGSEA_normalized" for plotting with heatmapEnrichment()? I have tried both approaches and the plots look quite different, so I am having trouble understanding why there is such a discrepancy.

2. Additionally, I am running FindAllMarkers() to statistically test the differences we observe on our heatmapEnrichment() plots. With the way I have it set up, FindAllMarkers() uses "escape.ssGSEA_normalized" whereas heatmapEnrichment() uses "escape.ssGSEA" with z-scaling. This has also presented some discrepancies between the visual representation and the differential pathway analysis results. Would it be appropriate to use the z-scaled ES from "escape.ssGSEA" in FindAllMarkers()?

Your advice would be greatly appreciated, thank you so much!

Best,
Anna

[ncborcherding]

Hey @ayin0510

Thanks for reaching out and good questions.

What is the difference between using "escape.ssGSEA" with z-scaling as opposed to "escape.ssGSEA_normalized" for plotting with heatmapEnrichment()? I have tried both approaches and the plots look quite different, so I am having trouble understanding why there is such a discrepancy.

The plots should look different - they are plotting different values.

• z-scaling is converting the calculated enrichment score into the standard score based on the distribution of the data, using the mean and standard deviation of the distribution. This is done across all cells and gene sets and either a mean or median is used to visualize. This allows for a similar scale for all gene sets, allowing for the heatmap to not look funky.
• "escape.ssGSEA_normalized" is the output of performNormalization() which is what is invoked if you are using ```runEscape(,normalize = TRUE). This step is taking a scaling factor per cell and using it to normalize the enrichment score to the number of features of the cell or the default is actually scaling at a per cel and gene set level using the number of genes expressed in each gene set.

Additionally, I am running FindAllMarkers() to statistically test the differences we observe on our heatmapEnrichment() plots. With the way I have it set up, FindAllMarkers() uses "escape.ssGSEA_normalized" whereas heatmapEnrichment() uses "escape.ssGSEA" with z-scaling. This has also presented some discrepancies between the visual representation and the differential pathway analysis results. Would it be appropriate to use the z-scaled ES from "escape.ssGSEA" in FindAllMarkers()?

I think this make sense that they would not nessecarily be consistent with each other because of the normalization step. If you are going to use "escape.ssGSEA_normalized" for the FindAllMarkers() , you should probably use the same assay for the heatmap to be consistent. You can try scale = TRUE and scale = FALSE and see what suits you, the scaling is not as important.

Hope that helps and let me know if you have any other questions.

Nick

[ayin0510]

Hi Nick,

Thanks so much for taking the time to address my questions! Using the z-scaled normalized enrichment scores in the heatmap resolved the discrepancies I had between heatmapEnrichment() and FindAllMarkers().

Anna