kmeansGeneset: Cluster gene-sets by enrichment profiles with k-means...

In bedapub/ribiosGSEA: Gene-Set Enrichment Analysis Tools in Ribios

Cluster gene-sets by enrichment profiles with k-means clustering, and select representative gene-sets by gene-set composition

Description

Cluster gene-sets by enrichment profiles with k-means clustering, and select representative gene-sets by gene-set composition

Usage

kmeansGeneset(
  enrichProfMatrix,
  genesetGenes,
  optK = pmin(25, floor(nrow(enrichProfMatrix)/2)),
  iter.max = 15,
  nstart = 50,
  thrCumJaccardIndex = 0.5,
  maxRepPerCluster = 10,
  metaClusterColumns = 1:ncol(enrichProfMatrix)
)

Arguments

enrichProfMatrix	A numeric matrix representing gene-set enrichment profile. Each row represent one gene-set and each column represent one enrichment profile, for instance a contrast in differential gene expression analysis. The values of the matrix represent enrichment of gene-sets, for instance enrichment score or absolute log10-transform p-values can be used. The row names are gene-set names.
genesetGenes	A list of character strings, each element being genes of a gene-set in the enrichProfMatrix. The names of the list must exactly match the row-names of enrichProfMatrix, namely the names of gene-sets in the same order.
optK	Integer, the number of initial clusters of gene-sets. Because one or more gene-sets may be selected from each gene-set cluster, the number of finally selected gene-sets is equal to or larger than optK.
iter.max	Integer, the maximum numbers of iterations allowed. This parameter is passed to kmeans.
nstart	Integer, how many random sets should be chosen to initialize cluster centers. This parameter is passed to kmeans.
thrCumJaccardIndex	Numeric, between 0 and 1, the threshold of cumulative Jaccard Index. The larger the value is, the more gene-sets will be selected from each cluster
maxRepPerCluster	Integer, maximum number of representative genesets per cluster. If NULL or NA, no limit is set.
metaClusterColumns	Columns used to cluster the clusters by their average enrichment profile. By default, all columns are used. This function performs k-means clustering of enrichment profiles of gene-sets. Within each cluster, we first identify the union set of unique genes covered any gene-set in the cluster, and then calculate Jaccard Index between genes in each gene-set and the union set. Gene-sets are sorted descendingly by the Jaccard Index, and the cumulative Jaccard Index is calculated. Among the sorted gene-sets, the gene-sets up to the position when the cumulative Jaccard Index exceeds thrCumJaccardIndex are selected (excluding redundant gene-sets). The geneset clusters are ordered by their average profiles - similar clusters are near to each other.

Value

A list:

• kmeans Result object returned by kmeans.

• genesetClusterData A data.frame with following columns: GenesetCluster, GenesetInd, GenesetName, JaccardIndex, CumJaccardIndex, IsRepresentative.

• repGenesets Character vector, gene-set names that are selected as representative gene-sets from each gene-set clsuter.

• gsCompOverlapSelInd Factor vector, indicating the gene-set clusters represented by each representative gene-set.

Examples

set.seed(1887)
profMat <- matrix(rnorm(100), nrow=20, 
    dimnames=list(sprintf("geneset%d", 1:20), sprintf("contrast%d", 1:5)))
gsGenes <- lapply(1:nrow(profMat), function(x) 
    unique(sample(LETTERS, 10, replace=TRUE)))
names(gsGenes) <- rownames(profMat)
kmeansGeneset(profMat, gsGenes, optK=5)

bedapub/ribiosGSEA documentation built on March 30, 2023, 3:26 p.m.