AlignK not aligning clusters across k

[RvV1979]

I am having trouble aligning K based on my slist (attached in RDS format here: (https://github.com/royfrancis/pophelper/files/12670714/slist.RDS.zip) )

my code

slist<-readRDS("slist.RDS")
plotQ(slist,returnplot = T, exportplot =F, imgoutput="join")

generated the plot below:

my code
plotQ(alignK(slist),returnplot = T, exportplot =F, imgoutput="join")
including cluster alignment generated the plot below:

As you can see, cluster alignment is solved within each k. However, it is nowhere near perfectly aligned across different k. Do you have some advice on how to improve this?

Thanks

[henni164]

Just wanted to bump this- I have encountered the same issue with alignK across K from admixture input.

My fix was to rename the headers in the qlist dataframes for subsequent Ks based on the cluster names from the smallest K... a bit manual but it did work, since the cluster colors match the cluster names when you have not done alignK().

In your case, you could swap the column names between Cluster4 + Cluster3 at K = 4

something like colnames(slist[[5]]) <- c("Cluster1","Cluster2","Cluster4","Cluster3")

Hopefully this will help you at least get the look you are going for..

Cheers

[RvV1979]

Thanks for your reply. I was specifically looking for a non-manual solution. In the end, I implemented a fix based on the fix_colours() function from in https://github.com/TCLamnidis/AdmixturePlotter but then applying it to an slist. This seems to work for me (I am not a coder, however, and there may still be much room for improvement).

#attempt to replicate fixcolours from AdmixturePlotter script
slistk<-tabulateQ(slist)$k
for (i in 2: length(slistk)){
  k<-slistk[i]
  prevk<-slistk[i-1]
  slistattributes<-attributes(slist[[i]]) # copy attributes
  cor_mat<-cor(slist[[i-1]],slist[[i]])
  component_order<-apply(cor_mat, 1,which.max) 
  while (any(duplicated(component_order)) ) {
    duplicates <- as.numeric(component_order[which(duplicated(component_order))])
    for (duplicate in duplicates){
      condition = T
      while (condition == T) {
        cor_mat[cor_mat==min(cor_mat[component_order==duplicate,duplicate])] = -100 # remove lowest-scoring duplicate by setting correlation to -100
        component_order<-apply(cor_mat, 1,which.max)
        if (duplicate %in% as.numeric(component_order[which(duplicated(component_order))]) == F) {condition = F} # keep going as long as duplicates remain
      }
    }
  }
  if (k != prevk ) {component_order<-c(component_order, setdiff(1:k, component_order)) } # add additional component when k is increased
  slist[[i]]<-slist[[i]][,component_order]
  attributes(slist[[i]])<-slistattributes #restore attributes
}