FindAllMarkers in Seurat and statistical problem

Question

[piotto]

I used to use FindAllMarkers in Seurat for finding the differentially expressed genes (DEGs) across all clusters. The default statistical testing that Seurat use to calculate the DEGs via Wilcoxon Rank Sum Test which is also known as Mann-Whitney U test. As we know this test is a non-parametrical test which compares two groups. Now, I am wondering when we are using FindAllMarkers and this test is utilizing Wilcox as the default test, are the results reliable? Should not it used the Kruskal-Wallis instead?

Answer 1

[Nicolas Rosewick]

check here : FindConservedMarkers vs FindMarkers vs FindAllMarkers Seurat

In summary it compares each cluster vs "all other clusters" so Mann-Whitney U test is suitable for this ( 1 vs 1 )

[piotto]

Thank you Nikolas for your answer. So, the argumentation you have is that because in FindAllMarkers, to find the DEGs, it compares one cluster to the rest, so the comparison is 1 vs 1, yes? I am wondering now, in DEGs calculation, if this is the case, why Kruskal-Wallis did not used in Seurat to stay in the safe side? Even, if the distribution is normal, we may use one-way ANOVA in this case.