Indiscpecies - multipatt and overcoming multi-comparrisons What is the best way to overcome the potential issue of type 1 error when doing an indicator analysis using the multipatt function in the indicspecies R package? 
For example, I have 9 site types, and 12 000 OTUs (species). Im interested in looking at which OTUs are strongly indicative of different site types. But with so many species/site group combinations I know I definitely need to account for potential of type 1 error. 
Do I need to correct p values for the no. of species/ or number of total comparrisons? (What is the best way to do this)or.. could I just change the significant threshold to p<0.01 ?  
Currently using this code in R: 
indisp = multipatt(DNA_10_OTU, DNA_10_OTU$Cluster, print.perm = TRUE, control = how(nperm=999))
indisp$sign
 A: I've been using the Benjamini & Hochberg p-value correction in p.adjust in the stats package.  I realized I needed to do enough permutations to be able to calculate a small enough p-value that would be significant after being corrected.  There is probably a way to actually calculate this, but so far I have increased the number of permutations until there was no change in the corrected p-values (ex. 10,000 to 100,000).  I don't think this counts as "p-hacking", but actually doing enough permutations to calculate exact probabilities.  
example code would look like this (I'm addicted to data.table, apologies if confusing)
library(data.table)
indisp<- multipatt(DNA_10_OTU, DNA_10_OTU$Cluster, print.perm = TRUE, control = how(nperm=999))
#extract table of stats
indisp.sign<-as.data.table(indisp$sign, keep.rownames=TRUE)
#add adjusted p-value
indisp.sign[ ,p.value.bh:=p.adjust(p.value, method="BH")]
#now can select only the indicators with adjusted significant p-values
indisp.sign[p.value.bh<=0.05, ]

I did stumble on this question because I am trying to determine how to use strassoc to calculate confidence intervals for each of the significant indicators as suggested by De Cáceres, M. and Legendre, P. (2009) when multipatt finds group combinations to be more significant than individual groups.  I don't have a satisfactory answer to that, but figured I'd mention it as it is a concern I currently have.  
