How to perform pairwise comparisons between three or more groups in R

[LeandroD94]

Dear ggloor, As requested, I continue here the topic from https://github.com/ggloor/ALDEx2_dev/issues/33 opened by the user carburator. I rewrite my question below, for clarity of the other users.

My aim is to use ALDEx2 to explore the pairwise differences between three groups (A vs B, B vs C, A vs C). To do so, I want to use the glm module as suggested in the vignette but I don't understand how to get EVERY pairwise comparison. Then, I post down there a short R script just to create a dataset that is similar to the one of my interest (aside from the abundances).

R Script

library(ALDEx2) sample_number<-30 obs_number<-50

Just an example dataset

feature_table<-NULL for(x in 1:obs_number){ new_obs_Abundance<-c( rnbinom(n=sample_number/3, size=2000, mu = 900), rnbinom(n=sample_number/3, size=2000, mu= 1800), rnbinom(n=sample_number/3, size=2000, mu=50)) new_obs_Abundance[sample(sample_number, size = sample_number/1.5)] <- 0 # a false zero inflation feature_table<-rbind(feature_table, new_obs_Abundance) } row.names(featuretable)<-paste0("Observation",1:obs_number) colnames(featuretable)<-paste0(rep("Sample"),1:sample_number)

building the sample data

Target_groups<-factor(c(rep("A",sample_number/3), rep("B",sample_number/3), rep("C",sample_number/3) ), level=c("A","B","C") ) Confounding_factor<-c(rep("S",sample_number/2),rep("R",sample_number/2))

using ALDEx2 glm module

mm <- model.matrix(~ Confounding_factor + Target_groups) # name of the sample data for last aldx <- aldex.clr(feature_table, mm, mc.samples=128, denom="all", verbose=T ) aldx2 <- glm.test <- aldex.glm(aldx, mm) aldx3 <- aldex.glm.effect(aldx, CI = F) aldx_final <- data.frame(aldx2,aldx3) head(aldx_final)

three groups results

p_val<-aldx_final$Target_groupsB.pval.holm # I know I should check the CI, but this approach allows to simplify my question Results_A_vs_B<-aldx_final[p_val<0.05] p_val<-aldx_final$Target_groupsC.pval.holm Results_A_vs_C<-aldx_final[p_val<0.05]

Then...

Assuming that what I wrote until now is correct, at this point I just want to get ALSO the results of the last comparison (B vs C) but I only have the ones against the reference level (A). How to do this? Should I to re-level the factor changing the reference level or can I use the estimates of the levels B and C? I have already tried to remove the intercept (0 + ) but I did not know how to continue.

Thank you so much for your time and support, Leandro

[natalia-rodilla]

Dear Greg Gloor and Leandro, I have a very similar situation. I have 7 groups and I want to make specific comparisons among them, I do not have a particular reference group because it is not applicable in my case. Would there be a way of specifying the contrasts (comparisons) of interest? Thank you very much in advance. Best regards, Pilar

[ggloor]

Hi Leandro and Pilar

In order to do this, I will need to integrate the multcomp function into the aldex.glm back end (see https://stackoverflow.com/questions/51702221/glm-in-r-give-all-comparisons) for how to do this.

This will be a fairly major re-write

[matteoramazzotti]

Hi everybody, same issue with me, need to assess pairwise a multi-group dataset. The multcomp function proposed by ggloor would fit perfectly for my data and in general I think that the aldex approach would greatly benefit from it in the scope of "competing" for accuracy with other tools currently used (and abused) Best, Matteo

[SaraBertorello]

Good morning everyone, It would be very useful to me to perform pairwise comparisons between three (or more) groups. I hope a way is found soon, Sara

[aimirza]

Hi Leandro and Pilar

In order to do this, I will need to integrate the multcomp function into the aldex.glm back end (see https://stackoverflow.com/questions/51702221/glm-in-r-give-all-comparisons) for how to do this.

This will be a fairly major re-write

This would be super useful!