Differences to topGO enrichment

[chrarnold]

I am having some questions about the enrichGO function, and its comparability with the results obtained from the topGO package. I'd appreciate if someone can comment on my questions, I think this is of general help:

1. Are there any methodological differences between the two packages? Which type of algorithm / test combination clusterProfiler'' corresponds best as compared totopGOwhen using it in the default mode with the Fisher statistic and theweight01` algorithm?
2. p-value adjustment: enrichGO returns p.adjust and qvalue, while topGO returns only raw p-values that, however, are claimed to "not be affected by the multiple testing problem because they are not independent due to the nestedness and correlation of neighboring GO terms". Details can be found here (https://datacatz.wordpress.com/2018/01/19/gene-set-enrichment-analysis-with-topgo-part-1/). The authors also note in the paper (https://academic.oup.com/bioinformatics/article/22/13/1600/193669). What is your take here, especially because you report adjusted p and q values?

Thank you!

[chrarnold]

When I look at a correlation plot of topGO p-values and clusterProfiler p-values (unadjusted), I see a wild mixture:

• Points exactly along the diagonal
• off-diagonal points in both directions
• strikingly, though, I see many (!) points for which the topGO p-value is 1 and the clusterProfiler p-value is something between 0 and 1, it looks essentially like a horizontal line spanning the whole 0-1 interval. From a first look, the terms these points correspond to seem to be rather generic, such as "DNA methylation or demethylation".