Suggest sample groups from metadata text

Exploratory helper for GEO series whose phenotype columns do not state the case/control group cleanly. Concatenates each sample's text metadata, computes Jaro-Winkler string distances and cuts an average-linkage hierarchical clustering into k groups. The result is written to a new cluster column; condition is not changed, because automatically derived labels must be checked by a person before they define a contrast.

Usage

mx_cluster_samples(study, k = 2, columns = NULL)

Arguments

study

A metaXpressStudy object.

k

Integer. Number of groups. Default: 2.

columns

Optional character vector of colData columns to use. Default: all character/factor columns except identifiers.

Value

The input study with colData(study)$cluster. A cross-tabulation against condition (if present) is printed.

Examples

counts <- matrix(rpois(60, 50), nrow = 10,
                 dimnames = list(paste0("G", 1:10), paste0("S", 1:6)))
meta <- data.frame(title = c(paste("healthy donor", 1:3),
                             paste("tumour patient", 1:3)))
if (requireNamespace("stringdist", quietly = TRUE)) {
  s <- mx_cluster_samples(metaXpressStudy(counts, meta))
  s$cluster
}
#> Assigned 6 samples to 2 clusters in 'cluster'. Inspect before using as condition labels.
#> [1] "cluster1" "cluster1" "cluster1" "cluster2" "cluster2" "cluster2"