post-hoc test, ANOVA

[BigOmics Analytics Team]

I have another question regarding the differential gene expression analysis concerning the meta-q values in the context of three contrasts. Are these values calculated from the post-hoc test after an ANOVA?

[Thorben S., 02.09.2021]

[BigOmics Analytics Team]

Hi Thorben,

In the differential expression module, Omics Playground only compares two contrasts and a post-hoc test is therefore not necessary, but you can create many of them. We never tests three (or more) contrasts in a single ANOVA test. This is to keep things simple. In the 'feature-set ranking' (under clustering), we do use ANOVA on multiple groups for the factors for calculating a single p-value per factor.

Ivo

[Thorben Sauer]

Dear Ivo,

thank you for your reply. I always thought that multiple testing of groups within a single "experiment setup" has to be corrected with the ANOVA and post-hoc, due to alpha-error accumulation, even when only two groups are compared to each other in a specific comparison, but multiple comparisons are made. Is this specifically unnecessary when using DeSeq2, edgeR and trend.limma?

Best

Thorben

[BigOmics Analytics Team]

Hi Thorben.

It's confusing because (I think) there are two types of "multiple testing" involved if you are doing ANOVA for many gene in more than two groups.

The first type, is that you are not testing 1 gene but many genes, like 20000. So here you must use multiple test correction to control the FDR.

The second type of multiple testing is if you have more than two groups (for a single gene). Then (strictly speaking) you need to do another multiple testing correction because you need to do a post-hoc test to see which groups are significant of all the group combinations.

As said, we don't do testing of multiple groups (only two groups), but of course we have multiple genes. so we need to correct for the first type of multiplicity. This is independent of DeSeq2, edgeR or LIMMA.

Hope this helps,

Ivo