fastwilcoxGMTall with Goterms instead of pathways

[MaximePolicarpo]

Dear All,

I had two questions concerning enrichment analysis and permulations.

1. With the function "fastwilcoxGMTall", is it correct to use a list of goterms and associated genes instead of pathways ?

2. I performed 1,000 permulations for a binary trait analysis using "getPermsBinary" and then computed the p-values using permpvalcor. This result in a data.frame with two columns: "permpval", and "permstats". However, there is no Rho values, which are only in the getPermsBinary results. I guess that such Rho values don't have much sense as they represent the Rho values between the gene rates are the permutated phenotype on the tree. If I want to report the results of RERconverge, is it correct to report the initial Rho value computed with "correlateWithBinaryPhenotype", but with the p-values and stats values corresponding to the permulations results ? Also, in order to perform an enrichment analysis after those permulations, is it correct to run "fastwilcoxGMTall" with the "permstats" values as input ? (The same question hold true for permulations with a continuous trait)

Thanks a lot for any help and guidance ! :D

All the best,

Maxime

[nclark-lab]

Hello Maxime,

1. It is perfectly fine to use a list of GO terms or any gene annotations the user wants to supply.
2. There is some disagreement in the group about ranking by perm p-values versus the initial Rho values. We will consult and get to you with best practices.

[MaximePolicarpo]

Dear Nathan,

Thanks a lot for your answer.

For the moment, here is what I do:

I compute the perm_stats and perm_pvalues based on 1,000 permulations. Then, I build a named vector with the perm_stats values (and gene names as names) and use this named vector in the fastwilcoxGMTall function.

To note, on the side, I also did pGLS to correlate the copy number in several gene families and my phenotype of interest. I then converted the R-squared values from pGLS to R values (sqrt(R2) * sign(b1))
Using the functions you provide in the "Permulation Walkthrough", I could also perform 1,000 permulations of my phenotype and recomputed these pGLS and those R values. I then computed the permulations pvalues and permulations statistics (just by mimicking the RERconverge::permpvalcor function). I am now using the computed permulations statistics to perform the enrichment analysis (Wilcoxon Rank-Sum enrichment) but I will wait for your final answer to be sure of what I am doing :). Knowing that I also don't know if you assessed the differences in the enrichment analysis if you perform such a Wilcoxon Rank-Sum enrichment or a more simple over-representation analysis with the "significant" genes as input and all the tested genes as background ?

Thanks a lot again for your precious help,

All the best,

Maxime