R loop until condition is met by changing input data - r

Question

I have a Rna-seq dataset (rows=samples, columns=genes) which goes into a clustering function. The genes are grouped into n clusters labelled with numbers where the ones belonging to cluster 0 are the non-clustered genes. The ones clustered goes back into clustering and again we get n clusters labelled with numbers with 0 being again the non-clustered genes. The process goes ahead until no further genes are classified into cluster 0. I need to loop into through this process in order to return the final clustering result along with the union of the genes belonging to cluster 0 at each iteration. I'm aware this could be done with either while or repeat. I had a try using repeat but is not working and the problem is that I have not really clear how to set this up properly.
#define my dataset
dat<-my_dataset
repeat{
#run the clustering
aa<-cluster(dat)
#if the cluster 0 has length 0 (no genes), assign clustering results to Mod and stop the loop
if (length(which(aa$colors==0))==0){
Mod<-aa
break
#otherwise, store the genes belonging to cluster 0 and set up a new dataset made up by clustered genes
} else{
noPass<-rownames(dat)[aa$colors==0]
dat<-dat[,which(aa$colors!=0)]
}
return(list(aa,noPass))
}
Any suggestion is really appreciated.

Answer 1

This sounds like a suitable problem for recursion:
define a function that will cluster the elements in its argument; if there are none left-out from the clustering, then return the elements, otherwise call itself using only the elements that are still in, and so on until there are none left-out.
cluster_until_none_left_out = function( elements ) {
aa=cluster(data[elements])
if (sum(aa$colors==0)==0) { return( list(aa,elements) ) }
else{ return( cluster_until_none_left_out(elements[aa$colors!=0]) ) }
}
It's a bit difficult to provide a full, running example without any real data, but here's a very simple one using mock data consisting of a vector of random integers, which a mock 'clustering' function splits into those that are greater than half the mean (the single 'cluster'), and those that are less than this (the left-out, 'unclustered' elements).
data=sample(1:100)
cluster = function(vec) {
answer=data.frame(vec)
answer$colors=ifelse(vec>mean(vec)/2,1,0)
return(answer)
}
sum(cluster(data)==0)
# shows that one round of clustering leaves some elements in cluster 0
initial_elements=1:100
clustering_elements=cluster_until_none_left_out(initial_elements)[[2]]
sum(cluster(data[clustering_elements])==0)
# 0 (now nobody left in cluster 0)
The final clustering is now returned together with the indices of the elements it uses. The 'left-out' elements (cluster 0) are all the rest.
cluster_0 = initial_elements[ -clustering_elements ]