R/HclustGenesHiCor.R

In compmedlab/MCbiclust: Massive correlating biclusters for gene expression data and associated methods

#' Find the most highly correlated genes using hierarchical clustering
#' 
#' Upon finding an initial bicluster with \code{FindSeed()} not all the genes
#' in the chosen geneset will be highly correlated to the bicluster. 
#' \code{HclustGenesHiCor()} uses the output of \code{FindSeed()} and 
#' hierarchical clustering to only select the genes that are most highly correlated
#' to the bicluster. This is achieved by cutting the dendogram produced
#' from the clustering into a set number of groups and then only selecting the
#' groups that are most highly correlated to the bicluster
#' 
#' @param gem Gene expression matrix with genes as rows and samples as columns
#' @param seed Seed of highly correlating samples
#' @param cuts Number of groups to cut dendogram into
#' @return Numeric vector of most highly correlated genes
#' @example example_code/example_corscore.R
#' @export

HclustGenesHiCor <- function(gem,seed,cuts){
  
    row.names(gem) <- seq_len(dim(gem)[1])
  
    gem.cor <- cor(t(gem[,seed]),use = 'pairwise.complete.obs')
    
    gem.dend <- as.dendrogram(hclust(dist(gem.cor)))
    gem.hclust <- hclust(dist(gem.cor))
    gem.cuts <- cutree(gem.hclust, k=cuts)
  
    hclust.genes.list <- lapply(seq_len(cuts),
                                function(x) which(gem.cuts == x))
 
    temp.fun <- function(i) CorScoreCalc(data.frame(gem)[hclust.genes.list[[i]],],seed)
    hi.cor.values <- vapply(X=seq_len(cuts),temp.fun, FUN.VALUE = numeric(1))
  
    cor.value <- CorScoreCalc(gem, seed) 
  
    return(unlist(hclust.genes.list[which(hi.cor.values > cor.value)]))
}